Intelligent fault diagnosis method for electro-hydraulic driving tilting mechanism based on redundancy

Abstract

The invention belongs to the field of intelligent fault diagnosis of an electro-hydraulic drive tilting rotorcraft structure, and particularly relates to an intelligent fault diagnosis method of an electro-hydraulic drive tilting mechanism based on redundancy. Comprising the steps that 1, three-redundancy multi-mode signals are collected and preprocessed, and the multi-mode signals comprise pressure, motor rotation, a tilt angle, servo valve displacement and current; 2, voting is carried out on the preprocessed three-redundancy multi-mode signals respectively; step 3, carrying out feature extraction on the multi-mode signal after voting; 4, performing first fault diagnosis based on a physical model and performing second fault diagnosis based on a data probability according to the extracted features and redundant null-space components of the multi-modal signals; and fusing, discriminating and outputting the first fault diagnosis result and the second fault diagnosis result.

Description

Technical Field

[0001] This invention belongs to the field of intelligent fault diagnosis of electro-hydraulic driven tilting rotor mechanisms, specifically relating to an intelligent fault diagnosis method for electro-hydraulic driven tilting mechanisms based on redundancy. Background Technology

[0002] Tiltrotor aircraft are a new type of aircraft that combines the vertical takeoff and landing capabilities of helicopters with the high-speed cruise advantages of fixed-wing aircraft. One of its core technologies is the rotor tilting mechanism, which controls the nacelle to switch between vertical and horizontal modes via an electro-hydraulic drive system. Since the tilting mechanism is a key functional component ensuring smooth transitions between flight modes, its reliability and safety are directly related to flight safety. To achieve extremely high reliability requirements, tilting actuation systems currently generally adopt a redundant architecture design. Under this architecture, how to quickly and accurately monitor and diagnose potential faults in the system is of paramount importance. Existing fault diagnosis methods mostly rely on threshold judgment of a single signal or residual analysis based on simplified mathematical models. These methods have the following shortcomings: 1. Information bias: A single sensor signal cannot fully reflect the complex operating state of the system. For fault modes with strong coupling, it is easy to produce missed or false alarms.

[0003] 2. Weak early fault detection capability: When the system experiences early performance degradation or minor faults, the characteristic changes of a single signal are not obvious, making it difficult for traditional methods to effectively capture them.

[0004] 3. Low level of diagnostic intelligence: Traditional methods rely on preset fault thresholds and fixed logic models, which are poorly adaptable to unknown or nonlinear fault modes, and the level of intelligence needs to be improved.

[0005] Therefore, how to effectively integrate multimodal information from different types of sensors to achieve rapid, accurate, and intelligent fault diagnosis of electro-hydraulic driven rotor tilting mechanisms is a key technical challenge for improving the safety and reliability of tiltrotor aircraft. Summary of the Invention

[0006] Purpose of the invention: To provide an intelligent fault diagnosis method for electro-hydraulic driven tilting mechanisms based on redundancy.

[0007] Technical solution: A method for intelligent fault diagnosis of electro-hydraulic driven tilting mechanisms based on redundancy, comprising: Step 1: Acquire and preprocess the multimodal signals of the triple redundancy, including pressure, motor resolver, tilt angle, servo valve displacement, and current. Step 2: Vote on the preprocessed triple-redundant multimodal signals respectively; Step 3: Extract features from the multimodal signals after voting; Step 4: Based on the characteristics of the extracted multimodal signals and redundant null space components The system performs a first fault diagnosis based on a physical model and a second fault diagnosis based on data probability; the results of the first and second fault diagnoses are then fused and output.

[0008] Furthermore, step one specifically includes: Step 11: Sampling at a uniform rate Synchronously sample all signals from all channels; Step 12: Denoise and normalize the synchronously sampled signal; Step 13: Calculate the short-time statistics of the normalized signal, including the short-time median and short-time variance.

[0009] Furthermore, step two specifically includes: Step 21: Calculate the median baseline based on the normalized signal, short-time median, and short-time variance; Step 22: Calculate the short-time confidence level for each channel based on the short-time variance. : ,in, This represents the minimum short-time variance. It is a constant; Calculate the residual consistency suppression term based on the median baseline and the normalized signal. : ,in, To control the intensity of deviation suppression, The signal after normalization The median baseline; Calculate the normalized confidence level based on the short-time confidence level and the residual consistency suppression term. : ,in, ,and ; Step 23: Utilize normalized confidence The normalized signal is used to calculate the reliable measurement sequence for each mode as the voting output. ,in, .

[0010] Furthermore, step three specifically involves: Step 31: Calculate the redundant zero-space components of the control input of the tilting mechanism; Step 32: Extract time-domain / frequency-domain features from each modal signal after voting; Step 33: Based on the time-domain / frequency-domain features, redundant null spatial components, and normalized confidence, perform fusion to obtain low-dimensional fused features. .

[0011] Furthermore, step four specifically includes: Step 41: For low-dimensional features Redundant null space components Sliding mode observations were performed to obtain the sliding mode quantity. ; Step 42: Utilize the overtorque sliding mode observer and sliding mode quantity Estimate fault parameters : Step 43: Input the discriminant network CNN-LSTM, output the data discriminant probability vector. ; Step 44: Estimate the observer Mapped to physical probabilities and normalized to ,Will Fusion ; according to Determine the type of fault .

[0012] Furthermore, step 41 specifically includes: ,in, For design vectors.

[0013] Furthermore, in step 42, ,in, , .

[0014] Furthermore, in step 44, ,in, Adaptive weights , ,in For average credibility, The equilibrium constant is adaptively adjusted.

[0015] Furthermore, in step 44, .

[0016] Beneficial effects: 1. Comprehensiveness and accuracy of diagnosis: By fusing information from multiple modal sensors such as pressure, speed, and displacement, and extracting multi-dimensional features, the limitations of a single information source are overcome, enabling a more comprehensive characterization of the system state and significantly improving the accuracy and robustness of fault diagnosis.

[0017] 2. Early fault detection capability: By combining time domain, frequency domain and time-frequency domain analysis, and capturing transient features, the method becomes more sensitive to early performance degradation and weak fault signals of the system, enabling early warning and early diagnosis of faults.

[0018] 3. Highly intelligent diagnosis: By introducing an observer and discriminant network (CNN-LSTM), it can automatically learn complex fault characteristics from massive amounts of data without relying on precise mathematical models and cumbersome manual rule setting. It has stronger diagnostic capabilities and adaptability for unknown faults in nonlinear and time-varying systems.

[0019] 4. Improved diagnostic reliability: The system combines model-based and AI-based diagnostic methods, allowing for mutual verification and complementarity. This effectively reduces the false alarm and false negative rates of a single diagnostic model, thereby enhancing the reliability of the overall diagnostic decision. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. The drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 A hybrid intelligent fault diagnosis method architecture for redundant electro-hydraulic driven rotor tilting mechanisms; Figure 2 This is the result of the fault diagnosis test; Figure 3 This is a confusion matrix for fault diagnosis test results. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] The features and illustrative embodiments of various aspects of the present invention will now be described in detail. Numerous specific details are set forth in the following detailed description to provide a thorough understanding of the invention. However, it will be apparent to those skilled in the art that the invention may be practiced without requiring some of these specific details. The following description of embodiments is merely intended to provide a better understanding of the invention by illustrating examples of the invention. The invention is by no means limited to any specific setups and methods set forth below, but covers any improvements, substitutions, and modifications to structures, methods, and devices without departing from the spirit of the invention. Well-known structures and techniques are not shown in the drawings and the following description to avoid unnecessarily obscuring the invention.

[0024] In the description of this invention, it should be noted that the directions or positional relationships indicated by terms such as "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer" are based on the directions or positional relationships shown in the accompanying drawings and are only for the convenience of describing and simplifying the invention, and should not be construed as limiting the invention. Furthermore, the use of ordinal numbers (e.g., "first and second," etc.) is for distinguishing objects and is not limited to this order, and should not be construed as indicating or implying relative importance.

[0025] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly, encompassing both direct connection and indirect connection via an intermediate medium. Those skilled in the art can understand the specific meaning of these terms in this invention based on the specific circumstances.

[0026] It should be noted that, unless otherwise specified, the embodiments of the present invention and the features thereof can be combined with each other, and the various embodiments can be referenced and cited in each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0027] The present invention will be further described in detail below with reference to the embodiments and accompanying drawings, but the embodiments of the present invention are not limited thereto.

[0028] This invention proposes a hybrid intelligent fault diagnosis method for electro-hydraulic driven rotor tilting mechanisms based on redundancy. This method integrates multi-source heterogeneous sensor information such as pressure, rotational speed, and displacement under a triple-redundant architecture, combined with time-domain, frequency-domain, and time-domain-frequency-domain feature extraction and artificial intelligence diagnostic models, to achieve accurate and real-time monitoring, status assessment, and fault diagnosis of key components of the tilting mechanism.

[0029] like Figure 1-3 A hybrid intelligent fault diagnosis method for electro-hydraulic driven rotor tilting mechanisms based on redundancy mainly includes the following steps: Step 1: Multimodal signal acquisition and preprocessing Objective: To acquire synchronous, multi-channel redundant raw data and output a denoised, normalized, and time-aligned dataset for subsequent voting and feature extraction.

[0030] Sub-step 1.1 Sensor layout and synchronous sampling Input: Sensor hardware (displacement / angle / pressure / current / velocity, etc., three-channel redundant configuration); sampling time is consistent with controller bus time.

[0031] Processing: Based on a uniform sampling rate (200 Hz) Synchronous sampling of all channels; timestamp alignment; packet loss / jitter flags.

[0032] Output: in,

[0033]

[0034] Sub-step 1.2 Denoising and Normalization enter: .

[0035] Processing: Denoise each channel. (represented as wavelet thresholding denoising) and For min-max normalization: .

[0036] Output: Denoising and normalized synchronization signal .

[0037] Sub-step 1.3 Short-time statistics calculation (preparing for weighting and anomaly screening in step 2) enter: Window width (e.g., 0.5s–2s).

[0038] Processing: Calculate short-time median, short-time variance, etc.

[0039] Output: and short-time statistics set

[0040] Transmission Note: These short-term statistics serve as the channel confidence level in step 2. Input.

[0041] Step Two: Triple Redundancy Signal Voting and Verification Objective: To obtain a single reliable measurement of each key physical quantity using a triple redundancy strategy, while simultaneously outputting the confidence level / anomaly flag for each channel for use in step 3. The voting here is no longer a simple median / majority vote, but rather an "adaptive confidence-weighted voting" system.

[0042] Input (from step 1) Denoising and normalization of the synchronization signal Short-time statistics set

[0043] Sub-step 2.1 Median baseline calculation (traditional baseline) Processing: Take the median value of each physical quantity as the initial baseline: .

[0044] Output: Baseline .

[0045] Sub-step 2.2 Channel Short-Time Confidence Calculation (Innovation) deal with: Short-time confidence level (approximately the reciprocal of SNR):

[0046] illustrate: This is the minimum short-time variance of sub-step 1.3. Preventing division by zero (recommended) – ).

[0047] Residual consistency suppression term (attenuates abrupt changes that deviate from the baseline):

[0048] illustrate: Control the strength of the suppression of deviation (empirical 1–10, adjusted according to dimensions).

[0049] Normalized confidence level (weight):

[0050] illustrate: ,and .

[0051] Output: Weight per channel (Real-time credibility).

[0052] Sub-step 2.3 Voting results and anomaly markers Processing: Reliable measurement of weighted voting output:

[0053] If a certain channel (For example )or If it is, it will be marked as a "suspected fault" and recorded in the exception queue.

[0054] Output (passed to step 3): Reliable measurement sequence after voting ; Channel normalization reliability ; Step 3: Multidimensional Feature Extraction and Fusion (Use the output of Step 2 as a required input) Objective: To transform voting measurements, raw modes, and redundant driving information into low-dimensional features with strong discriminative power. And explicitly calculate the redundant null space components. As a key feature.

[0055] Input (required) Voting Measurement (From step 2) Raw / Preprocessed Signals (From step 1) Channel reliability (From step 2) Control input With the drive allocation matrix (System knowledge) Sub-step 3.1 Calculate the redundant null space components (key innovation) Given a simplified linear driving force distribution model:

[0056] Define the generalized inverse With null projection matrix:

[0057] Actual calculation of redundancy vector (if the control input is known) ):

[0058] Feature extraction includes .

[0059] Sub-step 3.2 Time / Frequency Domain Extraction (Standard + Quantities to be Output) Handling: and extract: Time domain: mean, variance, drift rate ; Frequency domain: Short-time Fourier spectrum, in-band / out-of-band energy ratio ; Output: Time / Frequency Feature Set .

[0060] Sub-step 3.3: Fusion and Dimensionality Reduction (Innovation: Attention-AE Associates the Weights from Step 2) enter: .

[0061] Processing: Constructing feature vectors: .

[0062] First, calculate the initial attention value (normalized):

[0063] illustrate: This is an adjustable scaling factor (affecting both magnification and smoothing reliability); if features are grouped by channel, a corresponding scaling factor can be applied to each group. .

[0064] Weighting the group features at the input of the autoencoder: (According to block multiplication).

[0065] Encoding dimensionality reduction using an autoencoder with attention mechanism: , This is the encoder. The encoder is given by an empirical formula.

[0066] Output (passed to step 4): Low-dimensional fused features Redundant components Confidence level .

[0067] Step 4: Intelligent Fault Diagnosis Based on Hybrid Model Purpose: Based on and The system locates the fault, identifies its type, estimates its parameters, and outputs its health status.

[0068] Input (from step 3) Low-dimensional features Redundant null space components Credibility

[0069] Sub-step 4.1 Integrating residual construction and sliding modulus design (for the observer) Processing: Define the fusion residual / sliding modulus

[0070]

[0071] in Design vectors (set based on experience).

[0072] Output: Sliding modulus .Should Switching / nonlinear correction terms used to drive the observer.

[0073] Sub-step 4.2 Physical parameter estimation: Overtorque sliding mode observer (SM-STE) Processing / Innovation: Online estimation of fault parameter vectors using SM-STE (Leakage rate) Friction gain offset Effective delay (etc.). The observer is:

[0074] select satisfy A function that is greater than the upper bound of the system disturbance or the upper bound of the estimation error disturbance. Used to adjust the convergence rate.

[0075] , .

[0076] Output: Fault parameter estimation And its confidence interval (estimation error bound).

[0077] Sub-step 4.3 Data-driven classifier output Processing: Put Input discriminative network (CNN-LSTM): Output data discriminative probability vector The network can be fine-tuned online (semi-supervised) to adapt to aging and changes in operating conditions.

[0078] Output: .

[0079] Sub-step 4.4 Physical-Data Probabilistic Fusion Decision Processing: Physical probability mapping: estimating the observer Mapped to physical probabilities (using Mahalanobis distance kernel function): After normalization, we get Estimate the observer Mapping to physical probability distribution Then merge:

[0080] Adaptive weights Based on channel credibility Sure( Follow (Increase as needed), the fusion weight can be set to ,in For average credibility, The equilibrium constant is adaptively adjusted.

[0081] Output: Fusion probability With judgment Determine the fault type.

[0082] The key innovation of this invention lies in: 1. High reliability: Through an adaptive credibility voting mechanism and zero-space redundancy quantization, the loss failure and noise interference of a single sensor are effectively filtered out, which greatly improves the quality of input data and the reliability of diagnostic decisions.

[0083] 2. High sensitivity: The credibility-driven attention fusion mechanism integrates multimodal information such as pressure, rotation speed, displacement, and angle, and performs deep feature mining from the time domain, frequency domain, and time-frequency domain, which can capture early and weak fault signs that are difficult to detect by traditional methods.

[0084] 3. High Accuracy: Utilizing a hybrid intelligent diagnostic approach, it effectively integrates the physical and data domains to automatically learn the complex mapping relationships between fault modes and multiple features. This overcomes the dependence on precise system modeling of mathematical model-based methods and the dependence on data volume of deep learning, resulting in a higher diagnostic accuracy rate. 4. Wide diagnostic scope: It can effectively diagnose various faults in key components such as servo valves, hydraulic motors, and sensors, providing technical support for system condition monitoring and predictive maintenance.

[0085] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for intelligent fault diagnosis of an electro-hydraulic driving tilting mechanism based on redundancy, characterized in that, The application relates to a three-redundancy multi-modal signal fusion method and device. Step one: collecting three-redundancy multi-modal signals and carrying out pretreatment, wherein the multi-modal signals include pressure, motor rotation, tilt angle, servo valve displacement and current; Step two: respectively voting the pretreated three-redundancy multi-modal signals; Step three: carrying out feature extraction on the voted multi-modal signals; Step four: according to the features of the extracted multi-modal signals and the redundant null space components , the first fault diagnosis is based on a physical model, and the second fault diagnosis is based on data probability; the first fault diagnosis result and the second fault diagnosis result are fused and discriminated to output.

2. The method of claim 1, wherein, Step one, specifically comprising: Step 11: Synchronized sampling at a uniform sampling rate Synchronized sampling of all signals for all channels; Step 12: carrying out denoising and normalization on the signals after synchronous sampling; Step 13: calculating short-time statistics of the normalized signals, wherein the short-time statistics include short-time median and short-time variance.

3. The method of claim 2, wherein, Step two, specifically comprising: Step 21: calculating a median baseline according to the normalized signals and the short-time median and short-time variance; Step 22: Calculate the short-time confidence of each channel according to the short-time variance : where, is the minimum value of the short-time variance, is a constant; According to the median baseline and the normalized signal, a residual consistency suppression term is calculated : , wherein is a control of the suppression strength to the deviation, is the normalized signal, is the median baseline; The normalized confidence is calculated according to the short-time confidence and the residual consistency suppression term : wherein , and ; Step 23: Utilize Normalized Confidence and the normalized signal, compute a reliable measurement sequence for each modality as the voting output: wherein .

4. The method of claim 3, wherein, Step three, specifically comprising: Step 31: calculating a redundancy null space component of the control input of the tilt mechanism; Step 32: carrying out time domain / frequency domain feature extraction on the voted multi-modal signals; Step 33: fusing to obtain low-dimensional fusion features according to the time domain / frequency domain features, the redundant null space components, and the normalized confidence .

5. The method of claim 4, wherein, Step four, specifically comprising: Step 41: Projecting the low-dimensional features , the redundant null-space components Performing a sliding mode observation to obtain the sliding mode quantities ; Step 42: Estimate the fault parameters using the super-torque sliding mode observer and the sliding mode quantity :​ Step 43: input discriminative network CNN-LSTM, output data discriminative probability vector ; Step 44: Observer estimates are mapped to physical probabilities and normalized to , are mapped to physical probabilities and normalized to , are mapped to physical probabilities and normalized to ; According to determining a fault type .

6. The method of claim 5, wherein, Step 41: specifically comprising: wherein, is the design vector.

7. The method of claim 6, wherein, In step 42, wherein , .

8. The method of claim 7, wherein, In step 44, wherein, Adaptive weights , where is the average credibility, is a balancing constant adaptive adjustment.

9. The method of claim 8, wherein, In step 44, 。

Images