Multi-information fusion hydraulic cylinder internal leakage fault diagnosis method

Abstract

The invention discloses a multi-information fusion hydraulic cylinder internal leakage fault diagnosis method, and relates to the technical field of hydraulic cylinder fault diagnos.The hydraulic cylinder internal leakage fault diagnosis method comprises the steps that firstly, a hydraulic cylinder heavy load, time-varying and nonlinear motion curve and a test system based on a multi-source sensor are constructed, and internal leakage fault evolution under the actual working condition is simulated; and secondly, developing a hybrid deep learning semi-supervised feature fusion framework based on a convolutional auto-encoder, a multi-head attention mechanism, residual connection and a bidirectional sequential network, and enhancing the characterization learning ability of a multi-source signal in a complex and multi-noise environment through omnibearing deep feature decoupling. And finally, feature extraction and fusion are realized by using the displacement error and the pressure signal, and fault diagnosis is completed. Experiments show that compared with a comparison model, the diagnosis accuracy of the provided model and feature fusion strategy is improved by 3.95%; under-3 decibel noise, the accuracy is improved by 2.33%, and the scheme has high accuracy, reliability and generalization performance.

Description

Technical Field

[0001] The present invention relates to the technical field of hydraulic cylinder fault diagnosis, and in particular to a multi-information fusion method for diagnosing internal leakage faults in hydraulic cylinders. Background Art

[0002] With the rapid development of industries such as aerospace, shipbuilding, and automotive, various types of hydraulic presses are finding widespread application due to their high efficiency, high precision, and high controllability. As the core actuator of hydraulic presses, the stability of hydraulic cylinders provides a reliable guarantee for production quality and safety. However, faced with complex working conditions and production environments, the types and probabilities of hydraulic cylinder failures are highly diverse, hidden, and random, making fault detection and diagnosis more challenging. Therefore, in-depth research on hydraulic cylinder fault diagnosis solutions is crucial for improving production efficiency, ensuring production safety, and promoting industry progress.

[0003] Hydraulic cylinder fault diagnosis technology was originally developed based on model-driven approaches. For example, a model-based fault detection and isolation scheme was employed in the literature to effectively isolate and diagnose faults in rudder servo systems used in ship navigation, leveraging models and physics. However, model-based diagnostic approaches rely heavily on expert systems and model building, making fault diagnosis more difficult as system complexity increases. With the advancement of sensors, research on hydraulic cylinder system fault diagnosis has advanced significantly in recent years.

[0004] Therefore, a multi-information fusion hydraulic cylinder leakage fault diagnosis method is proposed to solve the above problems. Summary of the Invention

[0005] In view of this, the technical problem to be solved by the present invention is to propose a multi-information fusion hydraulic cylinder internal leakage fault diagnosis method to solve the problem that the fault types and probabilities of hydraulic cylinders in the prior art are highly diverse, hidden and random.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a multi-information fusion hydraulic cylinder internal leakage fault diagnosis method, comprising the following steps: S1, System Construction: Construct a hydraulic cylinder heavy-load, time-varying and nonlinear motion curve and test system based on multi-source sensors to simulate the evolution of internal leakage faults under actual working conditions; S2, Model Development and Construction: Develop a hybrid deep learning semi-supervised feature fusion framework based on convolutional autoencoders (CAEs), multi-head attention mechanisms, residual connections, and bidirectional temporal networks (BiLSTMs). This framework enhances the representation learning capabilities of multi-source signals in complex and noisy environments through comprehensive deep feature decoupling. S3, feature fusion and diagnosis: use displacement error and pressure signal to realize feature extraction and fusion, and complete fault diagnosis.

[0007] Preferably, the specific implementation steps of the model described in S2 include: S2.1, Data Acquisition: Use the built-in displacement sensor directly associated with the system variables and the pressure sensor in the high-pressure chamber circuit of the hydraulic cylinder to collect sensor data under different leakage conditions; S2.2, Data Preprocessing: A closed-loop displacement control scheme is used to make the displacement data time-varying and nonlinear. The displacement data and theoretical displacement data are processed to obtain more stable motion deviation data. At the same time, pressure data and motion displacement deviation data are extracted according to the motion cycle and divided into training data and test data using a sliding window scheme to prevent data leakage and provide a guarantee for the subsequent establishment of an accurate model. S2.3, Autoencoder Model and Feature Extraction: A dual-channel convolutional autoencoder (CAE) is used to extract local features of displacement or pressure signals. A multi-head attention mechanism is used to capture global temporal dependencies. Residual connections are used to preserve original features. Finally, a bidirectional temporal sequence network (BiLSTM) is used for bidirectional temporal modeling and fully connected layer classification output, achieving efficient fusion of multi-source sensor data and fault feature extraction. S2.4, classification model and its training: Fuse the two sets of signal features from the fully connected layer, weight the fused features, use the rectified linear unit (ReLU) activation function for nonlinear transformation, and use the normalized exponential function (softmax) activation function to output the classification results.

[0008] Preferably, the autoencoder model and feature extraction described in S2.3 specifically include: S2.3.1, Dual-channel feature extraction: Construct two parallel convolutional autoencoder (CAE) channels and use the parallel convolutional autoencoder (CAE) structure to process multi-source sensor data. Each branch contains three convolutional layers and a maximum pooling layer to extract the local spatiotemporal features of displacement and pressure signals respectively. S2.3.2, Attention-enhanced Feature Fusion: The multi-head attention mechanism is used to capture the global temporal dependencies, while the residual connection preserves the original encoding features, solving the problem of long-range dependency information loss. S2.3.3, Bidirectional Time Series Modeling: Use a Bidirectional Time Series Network (BiLSTM) to perform bidirectional context modeling on the fusion features to enhance the temporal expression capability of fault features: S2.3.4, feature integration output: The fully connected layer flattens the multi-level features to complete the final representation.

[0009] Preferably, each channel in the dual channels described in S2.3.1 includes an encoder and a decoder. The encoder is composed of three convolutional layers and a maximum pooling layer, which can more effectively implement the task of extracting local features. The decoder decodes and reconstructs the features extracted after compression to achieve accuracy assurance of the convolutional autoencoder (CAE) model part.

[0010] Preferably, the S3 adopts a feature-level information fusion scheme for the fault diagnosis of hydraulic cylinder leakage under heavy load and time-varying speed conditions, and uses comprehensive feature information to effectively express the working state of the hydraulic cylinder. The specific formula is as follows:

[0011] Out error and Out pressure are the extracted displacement error and pressure signal output features respectively; Concat is a feature fusion connector; Input classifier Classifier input after feature fusion; den is the output of the fully connected layer; f den As the activation function of the fully connected layer, the rectified linear unit (ReLU) activation function is selected; W den and W out are the weights of the fully connected layer and the output layer respectively; b den and b out are the bias terms of the fully connected layer and the output layer respectively; out is the output layer output; f out is the activation function of the output layer, and the normalized exponential function (softmax) activation function is selected here.

[0012] Compared with the prior art, the present invention provides a multi-information fusion method for diagnosing hydraulic cylinder internal leakage faults, which has the following beneficial effects: This paper proposes a semi-supervised feature fusion fault diagnosis model for hydraulic cylinder internal leakage under heavy load and time-varying speed conditions, combining a convolutional autoencoder (CAE) with a multi-head attention mechanism, a residual network, and a bidirectional temporal sequence network (BiLSTM). This solution achieves high fault diagnosis results under various internal leakage signals and different noise environments. The main contributions of this research are as follows: (1) This paper designs a hydraulic cylinder fault simulation experiment based on heavy load and time-varying speed, which can more accurately simulate actual fault samples and better capture the internal leakage fault of the hydraulic cylinder in the real motion scene, and effectively realize the pressure signal and displacement error signal acquisition task under different internal leakage states.

[0013] (2) A semi-supervised feature fusion model based on a convolutional autoencoder (CAE) supplemented by a multi-head attention mechanism, a residual mechanism, and a bidirectional time series network (BiLSTM) module is proposed. This model achieves efficient feature fusion fault diagnosis tasks for long time series and heterogeneous signals.

[0014] (3) The diagnostic accuracy of fault data under complex noise conditions based on the experimental environment reached 100%, an increase of 2-20% compared to the control group. Considering the complexity of the production environment, the diagnostic effect of the data after adding different Gaussian noise to the original experimental environment noise can reach a minimum of 98.24%, and the high diagnostic accuracy can be maintained under different noise signals. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a flow chart of a multi-information fusion hydraulic cylinder internal leakage fault diagnosis method of the present invention; Figure 2 Schematic diagram of the information fusion strategy of the present invention; Figure 3 Schematic diagram of the basic structure of the model of the present invention; Figure 4 This is a schematic diagram of the operating principle of the system failure of the present invention; Figure 5 Schematic diagram of the main configuration and control model of the experimental equipment of the present invention; Figure 6 Schematic diagram of displacement error and pressure signal data sets for six types of faults in the present invention; Figure 7 Schematic diagram of data comparison of four evaluation indicators of seven types of models under the feature fusion strategy of the present invention; Figure 8 Schematic diagram comparing the four evaluation indicators of the single signal and feature fusion strategies of the present invention; Figure 9 This is a schematic diagram of the diagnosis accuracy of the model under different noise intensities of the present invention. DETAILED DESCRIPTION

[0016] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0017] Example 1, refer to the attached Figure 1 To the attached Figure 3 ; A multi-information fusion method for diagnosing internal leakage faults in hydraulic cylinders, comprising the following steps: S1, System Construction: Construct a hydraulic cylinder heavy-load, time-varying and nonlinear motion curve and test system based on multi-source sensors to simulate the evolution of internal leakage faults under actual working conditions; S2, Model Development and Construction: Develop a hybrid deep learning semi-supervised feature fusion framework based on convolutional autoencoders (CAEs), multi-head attention mechanisms, residual connections, and bidirectional temporal networks (BiLSTMs). This framework enhances the representation learning capabilities of multi-source signals in complex and noisy environments through comprehensive deep feature decoupling. The specific collection steps of this model are as follows: (1) Data acquisition: This solution uses a built-in displacement sensor directly associated with the system variables and a pressure sensor in the high-pressure chamber circuit of the hydraulic cylinder to collect sensor data under different leakage conditions.

[0018] (2) Data preprocessing: The system adopts a closed-loop displacement control scheme, which makes the displacement data time-varying and nonlinear, more in line with production operation requirements. At this time, to further improve the accuracy of the data, the displacement data and theoretical displacement data are processed to obtain more stable motion deviation data. At the same time, the pressure data and motion displacement deviation data are extracted according to the motion cycle and divided into training data and test data using a sliding window scheme to prevent data leakage and provide a guarantee for the subsequent establishment of an accurate model.

[0019] (3) Autoencoder model and feature extraction: Multi-source sensor data requires the establishment of a multi-channel input autoencoder model. By constructing two parallel convolutional autoencoder (CAE) channels, local feature information is extracted from the time series data of the two channels respectively. The encoder model consists of three convolutional layers and a maximum pooling layer, which can more effectively implement the task of extracting local features. The decoder decodes and reconstructs the features extracted after compression to achieve the accuracy guarantee of the convolutional autoencoder (CAE) model. In the second step, the multi-head attention mechanism first realizes the global feature attention of the front and back information of the time series and passes the information to the residual connection layer. The residual connection layer combines the feature information transmitted by the multi-head attention mechanism and the initial feature information provided by the encoder model to enhance the integrity of the information and ensure that key information is not missed. In the third step, a two-layer BiLSTM model performs bidirectional feature extraction on the information after the residual connection to ensure more comprehensive extraction of effective information. Finally, the fully connected layer flattens the feature information.

[0020] (4) Classification model and its training: First, the two sets of signal features output by the fully connected layer are fused in the classifier. At this point, the model freezes the parameters of each layer of the convolutional autoencoder (CAE) to reduce the running time of the classification model. Second, a weighted summation is performed on each node through the fully connected layer, and a rectified linear unit (ReLU) activation function is used for nonlinear transformation. Finally, the normalized exponential function (softmax) activation function is used to output the classification result.

[0021] S3, Feature Fusion and Diagnosis: Displacement error and pressure signals are used to extract and fuse features and perform fault diagnosis. A feature-level information fusion scheme is used to diagnose hydraulic cylinder leakage faults under heavy load and time-varying speed conditions. Comprehensive feature information is used to effectively express the working status of the hydraulic cylinder. Based on this, after the feature extraction module is trained, its parameters are frozen. The classification model first fuses the features extracted from the two sensor signals and then uses a fully connected layer and activation function to complete the classification task. The specific formula is as follows:

[0022] Out error and Out pressure are the extracted displacement error and pressure signal output features respectively; Concat is a feature fusion connector; Input classifier Classifier input after feature fusion; den is the output of the fully connected layer; f den As the activation function of the fully connected layer, the rectified linear unit (ReLU) is selected as the activation function; W den and W out are the weights of the fully connected layer and the output layer respectively; b den and b out are the bias terms of the fully connected layer and the output layer respectively; out is the output layer output; f out is the activation function of the output layer. Here, the normalized exponential function (softmax) is selected as the activation function.

[0023] Example 2, based on the above specific implementation method is as follows, Figure 4 To the attached Figure 9 As shown: First, the fault type and system description; Simulate fault principle and motion process Figure 4As shown, under heavy load conditions, the rod chamber of the hydraulic cylinder in the load-port independent system is always high-pressure. Oil leakage for all fault types is from the rod chamber to the rodless chamber. Displacement closed-loop control is used during the downward movement of the movable crossbeam, enabling data collection of displacement accuracy errors and rod chamber pressure for varying leakage rates under variable speed conditions. During the return stroke of the movable crossbeam, a servo motor achieves constant flow control. The controller sets the start and end points of the motion, and the servo valves in the rod and rodless chamber circuits are fully open, minimizing the effects of unstable variables caused by valve port adjustment.

[0024] Considering the safety of personnel and experimental environment, an external throttle valve is used to simulate the leakage fault in the upper and lower chambers of the hydraulic cylinder. The specific formula is as follows: Q leak =k*A tv *

[0025] Where, Q leak is the flow rate leaking through the throttle valve, k is the empirical coefficient; A tv is the throttle valve opening area; is the pressure difference at both ends of the throttle valve (i.e. the pressure difference between the upper and lower chambers of the hydraulic cylinder).

[0026] Based on the pressure differential between the servo valves in the upper and lower chambers of the hydraulic cylinder and the throttle valve opening size in the neutral position, an orifice flow model was used to calculate the hydraulic cylinder leakage flow rate at six different throttle valve openings. Combined with the actual operating characteristics of different hydraulic cylinder fault states, the leakage rate and corresponding fault type are shown in Table 1.

[0027]

[0028] Table 1 (leakage amount and failure type) Second, the experimental platform is built, such as Figure 5 As shown; To verify the effectiveness of the proposed model, we built Figure 5 The experimental bench equipment shown in the figure mainly includes: hydraulic station, valve block platform, simulated leakage device, hydraulic press, control station and power cabinet.

[0029] The hydraulic station, consisting of a fuel tank, servo motor, accumulator, and vane pump, provides the system with hydraulic power. Independent control servo valves for the upper and lower chambers of the hydraulic cylinder are mounted on the valve block platform, enabling time-varying closed-loop displacement control of the hydraulic cylinder through closed-loop control. The test bench utilizes a four-column design, with a large-mass slider simulating heavy-load motion scenarios found in industrial production, enhancing the applicability of internal leakage fault diagnosis experiments under these heavy-load conditions. A simulated leakage device connects the upper and lower chambers of the hydraulic cylinder via a T-joint, throttle valve, and hose, enabling simulation experiments with varying leakage levels. The control station primarily consists of a host computer, an acquisition card, and an output card. Using the Simulink platform, it acquires real-time displacement and pressure data from the hydraulic cylinder and outputs closed-loop motion control signals. The power cabinet provides high voltage and power to the control station, system, and servo motor, ensuring the proper operation of the test bench. Table 2 lists the key hydraulic and electronic components involved in signal generation and acquisition, based on experimental requirements.

[0030]

[0031] Table 2 (Main components and key parameters of the test bench) Third, experimental data collection, such as Figures 6 to 9 As shown; Considering the operating state of a hydraulic press under heavy loads, the health of the hydraulic cylinder during the time-varying downward phase significantly impacts product quality and operator safety. To analyze the hydraulic cylinder's motion under complex operating conditions, a shock-free displacement curve was designed for the movable crossbeam's lowering process, enabling closed-loop motion displacement control. An open pump control scheme was implemented for the return stroke.

[0032] The actual displacement signal and pressure signal generate strong signal noise due to factory noise interference, electromagnetic interference of sensor connection lines, solenoid valve connection lines, servo valve connection lines, signal transmission lines and sensor accuracy problems. At the same time, the signal frequency of the displacement sensor and pressure sensor is limited by the acquisition card and its own conditions, and there is strong timing interference in the acquisition process. In addition, the hydraulic cylinder has a long working time and a long single-cycle input timing data, which increases the difficulty of signal acquisition, alignment and processing. Based on this, the signal acquisition frequency is set to 71Hz / s through the control module (Simulink), and six different types of leakage degree data are collected respectively. Each type of data collects 76 cycles in total, shortening the length of the single-cycle motion timing series. The collected displacement error and pressure signal are as follows Figure 6 As shown in the figure, with the increase of the leakage of the hydraulic cylinder, under the condition of fixed total cycle, the single cycle motion time gradually lengthens, resulting in an extension of the total time.

[0033] Fourth, result analysis; Comparison of experimental results of original data feature fusion strategy; Figure 7 The results reveal the diagnostic evaluation indicators for single signals using the proposed model, two semi-supervised models, and two traditional models, using experimental data extracted from both the original production environment and simulated production conditions. As shown in the figure, the semi-supervised models all achieved diagnostic accuracy exceeding 90%, with overall evaluation indicators exceeding those of the traditional models. Furthermore, using a feature fusion strategy, the proposed model achieved 100% diagnostic accuracy and various indicators, representing a 3.95% improvement in accuracy compared to the best comparison model. Traditional one-dimensional convolutional models struggle to achieve good fault diagnosis efficiency for hydraulic cylinder leakage data, which exhibits strong time-varying, nonlinear, and long-series characteristics. Compared to other methods, the proposed model demonstrates its superior diagnostic capabilities.

[0034] Comparison of experimental results between single signal and feature fusion strategies; At the same time, in order to explore the difference between single signal conditions and feature fusion strategy, the proposed model is used to perform fault diagnosis on displacement error data and lower cavity pressure data, respectively, and the diagnosis and classification results of the proposed model under the feature fusion strategy are compared. Figure 8 Comparison of four evaluation indicators for diagnosis of three models. Further analysis Figure 8 It can be seen that the diagnostic performance of pressure signal data, which is widely used in hydraulic cylinder fault diagnosis, is 3.59% lower than that of displacement error signal data, indicating that pressure signals have certain limitations in expressing fault information. Furthermore, it can be intuitively concluded that the diagnostic performance after feature fusion is 5.83% higher than the best diagnostic performance of a single signal.

[0035] Figure 8 This study revealed that feature fusion strategies are more effective at diagnosing faults than single-signal models. This feature fusion approach effectively leverages the correlation and complementarity of signals collected by different sensors, addressing the insufficient diagnostic accuracy of single-signal methods for both micro- and large-leak conditions in this experiment, effectively improving the accuracy of hydraulic cylinder leakage fault diagnosis. Furthermore, the proposed algorithm demonstrated strong fault diagnosis capabilities across diverse signal sources in this comparison, demonstrating its robustness.

[0036] Comparison of experimental results of data under different noise conditions; Figure 9The diagnostic performance of the proposed model and the comparison model under different signal-to-noise ratio (SNR) conditions is visually demonstrated. As shown in the figure, the proposed model maintains a high diagnostic accuracy of over 98% under various SNR conditions, further demonstrating the stability and universality of the proposed model under different operating conditions. Furthermore, the proposed model improves diagnostic accuracy by at least 2.33% compared to other models. The proposed model, along with the convolutional autoencoder (CAE), and the convolutional autoencoder (CAE)-bidirectional temporal network (BiLSTM) semi-supervised learning models, significantly outperformed the supervised learning models convolutional neural network local feature extraction to temporal feature modeling (CNN-BiLSTM) and local feature extraction (CNN) for complex hydraulic cylinder data diagnosis, further validating the effectiveness of the proposed semi-supervised models in hydraulic system fault diagnosis. Furthermore, the diagnostic accuracy of the feature fusion model significantly surpasses that of the fault diagnosis using a single sensor. The figure shows that the feature fusion model improves diagnostic accuracy by at least 10% under the same conditions.

[0037] This study proposes a semi-supervised feature fusion fault diagnosis model for hydraulic cylinder internal leakage under heavy load and time-varying speed conditions, combining convolutional autoencoders, multi-head attention mechanisms, residual networks, and bidirectional temporal networks. This approach achieves high fault diagnosis results under various internal leakage signals and different noise environments. The main contributions of this study are as follows: (1) This paper designs a hydraulic cylinder fault simulation experiment based on heavy load and time-varying speed, which can more accurately simulate actual fault samples and better capture the internal leakage fault of the hydraulic cylinder in the real motion scene, and effectively realize the pressure signal and displacement error signal acquisition task under different internal leakage states.

[0038] (2) A semi-supervised feature fusion model based on a convolutional autoencoder (CAE) supplemented by a multi-head attention mechanism, a residual mechanism, and a bidirectional temporal network (BiLSTM) module is proposed. This model achieves efficient feature fusion fault diagnosis for long time series and heterogeneous signals.

[0039] (3) The diagnostic accuracy of fault data under complex noise conditions based on the experimental environment reached 100%, an increase of 2-20% compared to the control group. Considering the complexity of the production environment, the diagnostic effect of the data after adding different Gaussian noise to the original experimental environment noise can reach a minimum of 98.24%, and the high diagnostic accuracy can be maintained under different noise signals.

[0040] It should be noted that the term "comprises" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article, or apparatus that includes a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0041] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A multi-information fusion method for diagnosing internal leakage faults in hydraulic cylinders, characterized in that the steps include: S1, System Construction: Construct a hydraulic cylinder heavy-load, time-varying and nonlinear motion curve and test system based on multi-source sensors to simulate the evolution of internal leakage faults under actual working conditions; S2, Model Development and Construction: Develop a hybrid deep learning semi-supervised feature fusion framework based on convolutional autoencoders, multi-head attention mechanism, residual connections, and bidirectional temporal networks. This framework enhances the representation learning capability of multi-source signals in complex and noisy environments through comprehensive deep feature decoupling. S3, feature fusion and diagnosis: use displacement error and pressure signal to realize feature extraction and fusion, and complete fault diagnosis.

2. The multi-information fusion hydraulic cylinder internal leakage fault diagnosis method according to claim 1, characterized in that: The S2 specifically includes: S2.1, Data Acquisition: Use the built-in displacement sensor directly associated with the system variables and the pressure sensor in the high-pressure chamber circuit of the hydraulic cylinder to collect sensor data under different leakage conditions; S2.2, Data Preprocessing: A closed-loop displacement control scheme is used to make the displacement data time-varying and nonlinear. The displacement data and theoretical displacement data are processed to obtain more stable motion deviation data. At the same time, pressure data and motion displacement deviation data are extracted according to the motion cycle and divided into training data and test data using a sliding window scheme to prevent data leakage and provide a guarantee for the subsequent establishment of an accurate model. S2.3, Autoencoder Model and Feature Extraction: A dual-channel convolutional autoencoder is used to extract local features of displacement or pressure signals. A multi-head attention mechanism is used to capture global temporal dependencies. Residual connections are used to preserve original features. Finally, a bidirectional temporal network is used for bidirectional temporal modeling and fully connected layer classification output, achieving efficient fusion of multi-source sensor data and fault feature extraction. S2.4, classification model and its training: Fuse the two sets of signal features from the fully connected layer, weight the fused features, use the rectified linear unit activation function for nonlinear transformation, and use the normalized exponential function activation function to output the classification results.

3. The multi-information fusion hydraulic cylinder internal leakage fault diagnosis method according to claim 1, characterized in that: The autoencoder model and feature extraction described in S2.3 specifically include: S2.3.1, Dual-channel feature extraction: Construct two parallel convolutional autoencoder channels and use the parallel convolutional autoencoder structure to process multi-source sensor data. Each branch contains three convolutional layers and a maximum pooling layer to extract the local spatiotemporal features of displacement and pressure signals respectively; S2.3.2, Attention-enhanced Feature Fusion: The multi-head attention mechanism is used to capture the global temporal dependencies, while the residual connection preserves the original encoding features, solving the problem of long-range dependency information loss. S2.3.3, Bidirectional Time Series Modeling: A two-layer bidirectional time series network is used to perform bidirectional context modeling on the fusion features to enhance the temporal expression capability of fault features: S2.3.4, feature integration output: The fully connected layer flattens the multi-level features to complete the final representation.

4. The multi-information fusion hydraulic cylinder internal leakage fault diagnosis method according to claim 1, characterized in that: Each of the dual channels described in S2.3.1 includes an encoder and a decoder. The encoder consists of three convolutional layers and a maximum pooling layer, which can more effectively implement the task of extracting local features. The decoder decodes and reconstructs the features extracted after compression to ensure the accuracy of the convolutional autoencoder model.

5. The multi-information fusion hydraulic cylinder internal leakage fault diagnosis method according to claim 1, characterized in that: The S3 adopts a feature-level information fusion scheme for the fault diagnosis of hydraulic cylinder leakage under heavy load and time-varying speed conditions, and uses comprehensive feature information to effectively express the working status of the hydraulic cylinder. The specific formula is as follows: Out error and Out pressure are the extracted displacement error and pressure signal output features respectively; Concat is a feature fusion connector; Input classifier Classifier input after feature fusion; den is the output of the fully connected layer; f den As the activation function of the fully connected layer, the rectified linear unit activation function is selected; W den and W out are the weights of the fully connected layer and the output layer respectively; b den and b out are the bias terms of the fully connected layer and the output layer respectively; out Output of the output layer; f out is the activation function of the output layer, and the normalized exponential function activation function is selected here.

Images