Ship water lubrication stern bearing fault diagnosis method and system based on multi-source information fusion

Abstract

The invention discloses a ship water-lubricated stern bearing fault diagnosis method and system based on multi-source information fusion. The method comprises the following steps: synchronously acquiring multi-source monitoring data of a ship water-lubricated stern bearing in different health states; constructing a multi-modal fault diagnosis model, wherein the multi-modal fault diagnosis model comprises a vibration signal feature extraction sub-network, a working condition parameter feature extraction sub-network, a feature fusion module and a classification output module; performing end-to-end training on the constructed multi-modal fault diagnosis model by using the multi-source training data set to obtain a trained fault diagnosis model; and inputting multi-source monitoring data collected in real time into the trained fault diagnosis model to obtain a comprehensive fault diagnosis result. According to the invention, the accuracy, reliability and early warning capability of fault diagnosis are obviously improved, and key technical support is provided for condition-based maintenance of the ship power device.

Description

Technical Field

[0001] This invention relates to the field of marine power plant condition monitoring and intelligent operation and maintenance technology, and in particular to a method and system for diagnosing faults in marine water-lubricated stern bearings based on multi-source information fusion. Background Technology

[0002] The water-lubricated stern bearing is a key component of the propulsion system, and its working condition directly affects the ship's operational safety and economic efficiency. Traditional fault diagnosis methods mainly rely on the analysis of vibration signals. For example, experienced engineers may judge by listening and touching, or use signal processing technology to extract the time-domain and frequency-domain characteristics of vibration signals, and then combine them with pattern recognition algorithms for diagnosis.

[0003] However, these methods have significant limitations: First, they heavily rely on expert experience, making automation and standardization difficult; second, the fault characteristic signals of water-lubricated stern bearings are often drowned out by strong background noise (such as water flow noise and engine noise), making it very difficult to extract sensitive features from vibration signals alone; finally, and most importantly, the occurrence and development of stern bearing wear, spalling, and other faults are not only related to mechanical vibration but also closely related to their working environment, especially the quality, pressure, flow rate, and temperature of the lubricating water, as well as the bearing's operating clearance. For example, poor water quality (such as high sand content or abnormal pH) will drastically accelerate bearing wear, while insufficient water pressure and flow rate will lead to poor lubrication and cooling, causing high-temperature seizing. Existing technologies lack synchronous monitoring and comprehensive analysis of these key operating parameters, resulting in inaccurate diagnostic results, delayed early warnings, and difficulty in achieving true predictive maintenance.

[0004] Therefore, there is an urgent need in this field for an intelligent fault diagnosis solution that can comprehensively utilize information from multiple sources to achieve earlier, more accurate, and more reliable diagnosis. Summary of the Invention

[0005] To address the aforementioned problems in existing technologies, this invention aims to provide a method and system for fault diagnosis of water-lubricated stern bearings in ships based on multi-source information fusion. This method can automatically and efficiently fuse vibration data with key operating parameters, deeply explore the intrinsic correlation between faults and multi-source information, thereby significantly improving the accuracy, reliability, and early warning capability of fault diagnosis.

[0006] To achieve the above objectives, the present invention provides the following solution:

[0007] This invention provides a fault diagnosis method for ship water-lubricated stern bearings based on multi-source information fusion, comprising the following steps:

[0008] S1. Multi-source data acquisition and preprocessing steps: Simultaneously acquire multi-source monitoring data of the ship's water-lubricated stern bearing under different health conditions. The multi-source monitoring data includes at least the vibration signal of the stern bearing, the radial clearance of the stern bearing, and the state parameters of the lubrication water system; preprocess and align the various types of data to construct a multi-source training dataset.

[0009] S2. Steps for constructing a multimodal deep learning model: Construct a multimodal fault diagnosis model, which includes:

[0010] A vibration signal feature extraction subnetwork is used to extract depth temporal features from vibration signals;

[0011] A subnetwork for extracting operating parameters is used to extract features from the radial clearance and lubrication water system state parameters;

[0012] The feature fusion module is used to fuse feature vectors from two sub-networks to form a joint feature representation;

[0013] The classification output module outputs the fault state classification result based on the joint feature representation;

[0014] S3. Model training steps: Use the multi-source training dataset to perform end-to-end training on the constructed multimodal fault diagnosis model to obtain the trained fault diagnosis model.

[0015] S4. Fault diagnosis steps: Input the real-time collected multi-source monitoring data into the trained fault diagnosis model to obtain a comprehensive fault diagnosis result.

[0016] Optionally, in S1:

[0017] The lubricating water system status parameters include at least one of the following: water temperature, water pressure, flow rate, and water quality; the water quality parameters include the acidity or alkalinity of the water.

[0018] The vibration signal is a multi-channel time-series signal;

[0019] The radial clearance and lubrication system status parameters are time-series signals that change over time or statistical values ​​within a certain time window.

[0020] Optional preprocessing of various types of data includes:

[0021] The vibration signal is denoised, various data are normalized, and the measured values ​​of other parameters are timestamped based on the sampling time of the vibration signal to form a synchronous multi-source data sample.

[0022] Optionally, the vibration signal feature extraction subnetwork is a one-dimensional convolutional neural network, whose input receives a preprocessed vibration signal sequence;

[0023] The operating condition parameter feature extraction subnetwork is a fully connected neural network, and its input end receives a feature vector composed of radial clearance and lubrication water system state parameters.

[0024] The feature fusion module uses a concatenation method to connect the feature vectors output by the two sub-networks.

[0025] Optionally, the one-dimensional convolutional neural network includes an attention mechanism module for weighting the feature channels of the convolutional layer output to enhance key fault-related features.

[0026] Optionally, in S3, the training objective of the model is to minimize the cross-entropy loss between the predicted fault category and the true label; a validation set is used for monitoring during training, and an early stopping strategy is employed to prevent overfitting.

[0027] The present invention also provides a fault diagnosis system for a ship water-lubricated stern bearing, comprising: a multi-source data acquisition module, a multi-modal fault diagnosis module, and a result display and early warning module;

[0028] A multi-source data acquisition module is used to simultaneously acquire multi-source monitoring data of the ship's water-lubricated stern bearing under different health conditions. The multi-source monitoring data includes at least the vibration signal of the stern bearing, the radial clearance of the stern bearing, and the state parameters of the lubrication water system. The module preprocesses and aligns the various types of data to construct a multi-source training dataset.

[0029] A multimodal fault diagnosis module is used to construct a multimodal fault diagnosis model. The multimodal fault diagnosis model includes: a vibration signal feature extraction sub-network for extracting deep temporal features from vibration signals; a working condition parameter feature extraction sub-network for extracting features from the radial clearance and lubrication water system state parameters; a feature fusion module for fusing feature vectors from the two sub-networks to form a joint feature representation; and a classification output module for outputting fault state classification results based on the joint feature representation. The constructed multimodal fault diagnosis model is trained end-to-end using the multi-source training dataset to obtain a trained fault diagnosis model.

[0030] The results display and early warning module is used to input the real-time collected multi-source monitoring data into the trained fault diagnosis model to obtain comprehensive fault diagnosis results.

[0031] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement a fault diagnosis method for a ship water-lubricated stern bearing based on multi-source information fusion.

[0032] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements a method for diagnosing faults in a ship's water-lubricated stern bearing based on multi-source information fusion.

[0033] The beneficial effects of this invention are as follows:

[0034] 1. Multi-source information fusion for more comprehensive and accurate diagnosis: This invention breaks through the limitations of single vibration analysis, and for the first time simultaneously monitors and deeply integrates radial clearance, which directly reflects the mechanical state of the bearing, and multiple parameters of the lubrication water system (water temperature, water pressure, flow rate, and water quality), which directly affect the bearing's service life. The model can learn the comprehensive representation of faults across different information dimensions, greatly improving the accuracy of identifying complex faults, especially the ability to identify early, subtle faults and soft faults such as poor lubrication.

[0035] 2. Strong anti-interference and generalization ability: By introducing operating condition parameters, the model can effectively distinguish between normal state fluctuations caused by changes in external operating conditions (such as increased load or sudden changes in water quality) and real fault symptoms, significantly reducing the false alarm rate and enabling the system to maintain stable and reliable diagnostic performance under different navigation conditions.

[0036] 3. End-to-end intelligent diagnosis without manual feature engineering: By leveraging the powerful feature extraction capabilities of deep learning models, the complex, highly specialized, and experience-dependent manual feature extraction steps are avoided. This enables end-to-end automated diagnosis from raw multi-source data to diagnostic results, lowering the technical threshold and improving diagnostic efficiency.

[0037] 4. Proactive early warning for predictive maintenance: Since changes in parameters such as slow increase in clearance and deterioration of water quality often occur before severe vibration signals, this invention can detect potential fault risks earlier by continuously monitoring and analyzing these early indicators, providing a longer early warning time for condition-based maintenance, effectively avoiding the occurrence of serious accidents, and ensuring the safety of ship navigation. Attached Figure Description

[0038] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, 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.

[0039] Figure 1 This is an overall flowchart of the fault diagnosis method described in the embodiments of the present invention;

[0040] Figure 2 This is a structural block diagram of the fault diagnosis system described in an embodiment of the present invention. Detailed Implementation

[0041] 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, and 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.

[0042] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0043] Example 1: This example provides a fault diagnosis method for ship water-lubricated stern bearings based on multi-source information fusion, such as... Figure 1 As shown, it specifically includes:

[0044] S1. Multi-source data acquisition and preprocessing steps: Simultaneously acquire multi-source monitoring data of the ship's water-lubricated stern bearing under different health conditions. The multi-source monitoring data includes at least the vibration signal of the stern bearing, the radial clearance of the stern bearing, and the state parameters of the lubrication water system; preprocess and align the various types of data to construct a multi-source training dataset.

[0045] In this step, a sensor system is installed on the stern bearing test bench and data is collected.

[0046] Sensor configuration: A vibration acceleration sensor (sampling frequency 25.6 kHz) is installed on the stern bearing housing; an eddy current displacement sensor is used to measure the radial clearance between the shaft and the bearing (sampling frequency 1 kHz); water temperature, water pressure, and flow sensors, as well as an online water quality (such as turbidity and pH) analyzer (sampling frequency 10 Hz) are installed on the bearing lubrication water inlet pipe.

[0047] Data Acquisition: Multiple health states of the bearing (e.g., normal, wear, spalling, foreign object intrusion) are set, and the test bench is run under different speeds and load conditions. Crucially, all sensors are synchronously triggered by the same data acquisition card, ensuring all data channels have consistent and accurate timestamps. Sufficient data duration is collected for each state and condition.

[0048] Data cleaning: Wavelet threshold denoising (e.g., db4 wavelet, 5-level decomposition) is used to remove noise from the vibration signal. Median filtering is applied to all continuous operating parameters (gap, water temperature, etc.) to smooth abnormal fluctuations.

[0049] Data normalization: Z-score standardization is performed on all dimensions of data (vibration signal amplitude, gap value, temperature, pressure, etc.) to make the mean 0 and the standard deviation 1, so as to eliminate the influence of dimensions.

[0050] Data alignment and sample generation: Using the timestamps of high-frequency vibration signals as a reference, low-frequency operating condition parameter data are linearly interpolated and resampled to the same time point. Subsequently, a sliding window method (window length 2048 points, overlap rate 50%) is used to segment the synchronized long-time series data, generating a large number of short samples. Each sample is assigned a corresponding health status label.

[0051] S2. Steps for constructing a multimodal deep learning model: Construct a multimodal fault diagnosis model, which includes:

[0052] A vibration signal feature extraction subnetwork is used to extract depth temporal features from vibration signals;

[0053] A subnetwork for extracting operating parameters is used to extract features from the radial clearance and lubrication water system state parameters;

[0054] The feature fusion module is used to fuse feature vectors from two sub-networks to form a joint feature representation;

[0055] The classification output module outputs the fault state classification result based on the joint feature representation;

[0056] S3. Model training steps: Use the multi-source training dataset to perform end-to-end training on the constructed multimodal fault diagnosis model to obtain a trained fault diagnosis model. The trained model will perform automatic diagnosis.

[0057] Model Input: Input the generated sample into the pre-trained multimodal deep learning model. The sample contains a 2048-point vibration signal sequence and a vector consisting of 6 operating parameters (interval mean, water temperature, water pressure, flow rate, turbidity, and pH value).

[0058] Model inference: The model automatically executes the following sub-steps internally:

[0059] Feature extraction: Vibration data are extracted for deep spatiotemporal features through a one-dimensional convolutional neural network (1D-CNN) subnetwork; operating parameters are extracted for nonlinear features through a fully connected network (FCN) subnetwork.

[0060] Feature fusion: The feature vectors output by the two sub-networks are concatenated to form a joint feature representation that integrates the mechanical state and the lubrication environment.

[0061] Fault classification: The fused features are passed through the final fully connected layer and Softmax classifier to calculate the probability distribution of the sample belonging to various fault types.

[0062] S4. Fault diagnosis steps: Input the real-time collected multi-source monitoring data into the trained fault diagnosis model to obtain a comprehensive fault diagnosis result.

[0063] The model completes the calculation and outputs a diagnostic result, such as: "Fault type: minor wear; confidence level: 85.2%". This result includes the specific fault classification and the degree of certainty with which the model made this judgment.

[0064] S5. Determine if an alert is needed & S6. Trigger the alert:

[0065] This is a decision-making and execution process.

[0066] The system reads the confidence level of the output and compares it with a preset warning threshold (e.g., 80%).

[0067] If the confidence level is below the threshold, the system is judged to be in a normal state or with high uncertainty. The system does not trigger an alarm, and the process returns to S1 to start the next round of continuous monitoring.

[0068] If the confidence level is higher than or equal to the threshold, the system determines that a fault has occurred and immediately executes S6 to trigger an early warning. The early warning actions include: displaying a red text and pop-up notification on the local human-machine interface, activating the audible and visual alarm, and sending the alarm information (including time, fault type, and confidence level) to the remote monitoring center via the communication network.

[0069] This completes one diagnostic cycle. The system will then cycle through the process from S1 to S6 to achieve continuous, real-time status monitoring and intelligent fault diagnosis of the ship's water-lubricated stern bearing.

[0070] Example 2: Figure 2 As shown in this embodiment, a fault diagnosis system for ship water-lubricated stern bearings based on multi-source information fusion includes:

[0071] Multi-source data acquisition module, multi-modal fault diagnosis module, and result display and early warning module;

[0072] A multi-source data acquisition module is used to simultaneously acquire multi-source monitoring data of the ship's water-lubricated stern bearing under different health conditions. The multi-source monitoring data includes at least the vibration signal of the stern bearing, the radial clearance of the stern bearing, and the state parameters of the lubrication water system. The module preprocesses and aligns the various types of data to construct a multi-source training dataset.

[0073] A multimodal fault diagnosis module is used to construct a multimodal fault diagnosis model. The multimodal fault diagnosis model includes: a vibration signal feature extraction sub-network for extracting deep temporal features from vibration signals; a working condition parameter feature extraction sub-network for extracting features from the radial clearance and lubrication water system state parameters; a feature fusion module for fusing feature vectors from the two sub-networks to form a joint feature representation; and a classification output module for outputting fault state classification results based on the joint feature representation. The constructed multimodal fault diagnosis model is trained end-to-end using the multi-source training dataset to obtain a trained fault diagnosis model.

[0074] The results display and early warning module is used to input the real-time collected multi-source monitoring data into the trained fault diagnosis model to obtain comprehensive fault diagnosis results.

[0075] Specifically, the sensor array is used to measure vibration, clearance, and water quality parameters of the water-lubricated stern shaft system.

[0076] Data acquisition and processing unit: Receives raw signals from various sensors (vibration, gap, water temperature, water pressure, flow rate, water quality) and converts them into digital signals. Ensures all sensor data has a consistent and accurate timestamp. Performs noise reduction, filtering, and standardization on the data, while also acting as a data buffer and temporary storage unit.

[0077] The diagnostic and interaction unit performs feature extraction, information fusion, and status classification on the processed data, ultimately outputting the specific fault type and its confidence level. Diagnostic results, real-time equipment parameters, and historical status trends are displayed intuitively in a visual format (such as real-time data curves, dashboards, health status indicator lights, and confidence level progress bars). Based on the diagnostic results (especially the confidence level), it compares them with preset thresholds and automatically makes early warning decisions. When a fault is determined to have occurred, it automatically triggers predetermined early warning actions, such as audible and visual alarms, pop-up warning boxes, and sending warning messages (SMS, email). Finally, it stores key diagnostic results, alarm logs, and raw data, and typically has network communication capabilities, allowing data to be uploaded to the cloud or a higher-level management system for remote monitoring and persistence.

[0078] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A ship water lubricated stern bearing fault diagnosis method based on multi-source information fusion, characterized in that, The method comprises the following steps: S1, multi-source data acquisition and preprocessing step: synchronously acquiring multi-source monitoring data of a ship water-lubricated stern bearing under different health states, the multi-source monitoring data at least comprising a vibration signal of the stern bearing, a radial clearance of the stern bearing, and a state parameter of a lubricating water system; preprocessing and sample aligning of various data, and constructing a multi-source training data set; S2, multi-modal deep learning model construction step: constructing a multi-modal fault diagnosis model, the multi-modal fault diagnosis model comprising: a vibration signal feature extraction subnetwork for extracting deep time sequence features from the vibration signal; a working condition parameter feature extraction subnetwork for extracting features from the radial clearance and the lubricating water system state parameter; a feature fusion module for fusing feature vectors from the two subnetworks to form a joint feature representation; a classification output module for outputting a fault state classification result based on the joint feature representation; S3, model training step: using the multi-source training data set to perform end-to-end training on the constructed multi-modal fault diagnosis model to obtain a trained fault diagnosis model; S4, fault diagnosis step: inputting real-time acquired multi-source monitoring data into the trained fault diagnosis model to obtain a comprehensive fault diagnosis result.

2. The ship water lubricated stern bearing fault diagnosis method based on multi-source information fusion according to claim 1, characterized in that, In the S1: The state parameter of the lubricating water system comprises at least one of water temperature, water pressure, flow rate, and water quality; The parameter of the water quality comprises water pH or turbidity; The vibration signal is a multi-channel time sequence signal; The radial clearance and the lubricating water system state parameter are time-varying time sequence signals or statistical values within a certain time window.

3. The ship water lubricated stern bearing fault diagnosis method based on multi-source information fusion according to claim 2, characterized in that, The preprocessing of various data comprises: noise reduction of the vibration signal, normalization of various data, and time stamp alignment of the measurement values of other parameters based on the sampling time of the vibration signal to form synchronous multi-source data samples.

4. The ship water-lubricated stern bearing fault diagnosis method based on multi-source information fusion according to claim 1, characterized in that: The vibration signal feature extraction subnetwork is a one-dimensional convolutional neural network, and the input end of the one-dimensional convolutional neural network receives a preprocessed vibration signal sequence; The working condition parameter feature extraction subnetwork is a fully connected neural network, and the input end of the fully connected neural network receives a feature vector spliced from the radial clearance and the lubricating water system state parameter; The feature fusion module adopts a splicing manner to connect the feature vectors output by the two subnetworks.

5. The ship water lubricated stern bearing fault diagnosis method based on multi-source information fusion according to claim 4, characterized in that, The one-dimensional convolutional neural network comprises an attention mechanism module for weighting the feature channels output by the convolutional layer to enhance key features related to faults.

6. The ship water lubricated stern bearing fault diagnosis method based on multi-source information fusion according to claim 1, characterized in that, In the S3, the training target of the model is to minimize the cross-entropy loss between the predicted fault category and the true label; a validation set is used for monitoring during the training process, and an early stopping strategy is adopted to prevent overfitting.

7. A ship water-lubricated stern bearing fault diagnostic system for implementing the method according to any one of claims 1-6, characterized by It comprises: a multi-source data acquisition module, a multi-modal fault diagnosis module, and a result display and early warning module; A multi-source data acquisition module is configured to synchronously acquire multi-source monitoring data of a ship water-lubricated stern bearing in different health states, the multi-source monitoring data at least including a vibration signal of the stern bearing, a radial clearance of the stern bearing, and a state parameter of a lubricating water system; each type of data is preprocessed and sample-aligned, and a multi-source training dataset is constructed; A multi-modal fault diagnosis module is configured to construct a multi-modal fault diagnosis model, the multi-modal fault diagnosis model including: a vibration signal feature extraction subnetwork configured to extract deep time sequence features from the vibration signal; a working condition parameter feature extraction subnetwork configured to extract features from the radial clearance and the lubricating water system state parameter; a feature fusion module configured to fuse feature vectors from the two subnetworks to form a joint feature representation; a classification output module configured to output a fault state classification result based on the joint feature representation; and an end-to-end training of the constructed multi-modal fault diagnosis model using the multi-source training dataset to obtain a trained fault diagnosis model; A result display and early warning module is configured to input real-time acquired multi-source monitoring data into the trained fault diagnosis model to obtain a comprehensive fault diagnosis result.

8. An electronic device, comprising: A computer readable storage medium stores a computer program, and a processor executes the computer program to implement a ship water-lubricated stern bearing fault diagnosis method based on multi-source information fusion according to any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, and a processor executes the computer program to implement a ship water-lubricated stern bearing fault diagnosis method based on multi-source information fusion according to any one of claims 1-6.

Images