Marine equipment life evaluation strategy optimization method and system based on deep learning

Abstract

The invention provides a ship equipment life evaluation strategy optimization method and system based on deep learning, and relates to the technical field of life evaluation strategy optimization, and the method comprises the steps: obtaining the monitoring data of ship equipment, carrying out the redundancy processing, information integration, compensation filling and expansion similarity analysis, and obtaining the key processing data of the ship equipment; a deep learning model is trained, degradation feature training analysis is carried out, and then a preliminary life prediction model is constructed; obtaining time scale degradation analysis data, performing degradation category analysis and judgment and degradation change analysis according to the time scale degradation analysis data, and obtaining performance inflection point judgment information; according to the method, multiple service life segments and segment optimization factors of the ship equipment are obtained, stage sub-models are obtained according to the service life segments in combination with the segment optimization factors, adjustment and updating are carried out, a final service life prediction model is obtained, then service life evaluation strategy optimization data are obtained, and accurate and intelligent evaluation of the service life of the ship equipment is achieved.

Description

Technical Field

[0001] This invention proposes a method and system for optimizing the life assessment strategy of shipboard equipment based on deep learning, which relates to the field of life assessment strategy optimization technology, specifically to the field of deep learning-based optimization of life assessment strategies for shipboard equipment. Background Technology

[0002] In the field of shipboard equipment operation and maintenance, traditional life assessment and maintenance strategies mainly rely on periodic inspections and predictive methods based on physical models. Periodic inspections carry the risk of over-maintenance or under-maintenance, are costly, and cannot predict sudden failures. On the other hand, predictive methods based on physical models heavily rely on precise mathematical models and material parameters, making it difficult to cope with the complex and ever-changing actual operating conditions and comprehensive environmental stresses of ships. They are also insensitive to early, subtle degradation characteristics and have limited predictive accuracy. Summary of the Invention

[0003] This invention provides a deep learning-based method and system for optimizing shipboard equipment lifespan assessment strategies to address the aforementioned problems: The present invention proposes a deep learning-based method and system for optimizing shipboard equipment lifespan assessment strategies. The method includes: S1. Collect shipboard equipment data to obtain shipboard equipment monitoring data, perform redundancy processing, information integration, compensation filling and expansion similarity analysis on the shipboard equipment monitoring data, and obtain key processing data of shipboard equipment. S2. Train a deep learning model using key processing data from shipboard equipment, perform degradation feature training and analysis, and then construct a preliminary life prediction model. S3. Obtain time-scale degradation analysis data based on the preliminary life prediction model, perform degradation category analysis and determination based on the time-scale degradation analysis data, perform degradation change analysis based on the degradation category determination information, and obtain performance inflection point determination information based on the degradation change analysis data. S4. Based on the performance inflection point determination information, obtain multiple life segments of shipboard equipment, perform output deviation analysis on the preliminary life prediction model, and then obtain segment optimization factors. Based on the life segments and the segment optimization factors, obtain stage sub-models and adjust and update them to obtain the final life prediction model, and then obtain life assessment strategy optimization data.

[0004] Further, S1 includes: Multi-source data collection is conducted on shipboard equipment to obtain multi-source data from shipboard equipment; Data collected from multiple shipboard devices and sources is converted to a unified format to obtain multi-source converted data. Time-series data of multi-source converted data is obtained by time-labeling the collection time. Redundant data is removed from the collected time-series data to obtain redundant processed data; Redundant processed data is classified and stored across multiple collection categories to obtain a category-integrated database; Data missing information is obtained by identifying missing data in the category-integrated database. The missing information in the data is compensated and filled using time series prediction methods to obtain key processing data for shipboard equipment.

[0005] Furthermore, the step of using time series prediction methods to compensate for and fill in missing data to obtain key processing data for shipboard equipment includes: Missing time-series data is obtained by collecting time-series data and combining it with time-missing information; By using long short-term memory networks to predict and fill in missing temporal data, integrated temporal data can be obtained. Spatial texture information is obtained by performing two-dimensional image transformation on temporally integrated data using the Gram angle field method; Spatial texture information is augmented and augmented similarity analysis is performed. Based on the similarity analysis data, key feature processing is carried out to obtain key processing data for shipboard equipment.

[0006] Furthermore, the process of augmenting the spatial texture information and performing augmented similarity analysis, followed by key feature processing based on the similarity analysis data, to obtain key processing data for shipboard equipment includes: New augmentation points are generated between adjacent pixels of spatial texture information using bicubic interpolation to obtain an updated dataset. Obtain the similarity between the updated dataset and the corresponding data of adjacent pixels to obtain expanded similarity data; The expanded similarity data is compared with a preset similarity threshold to obtain a similarity comparison result; The updated dataset is assessed for eligibility based on the similarity comparison results to obtain similarity assessment information; Based on the similarity determination information, key feature data is extracted from the updated dataset to obtain key processing data for shipboard equipment.

[0007] Further, S2 includes: Deep learning models are trained using key data processed by shipboard equipment. Degradation feature data of key processing data of shipboard equipment is obtained through convolutional and pooling layers of deep learning models; A health index model is trained based on the degradation feature data, and health index trajectory data is output based on the health index model. Nonlinear regression analysis was performed on the health index trajectory data, and a preliminary lifespan prediction model was constructed based on the nonlinear regression analysis data.

[0008] Further, S3 includes: Based on the preliminary life prediction model and historical equipment operation data, degradation data analysis was performed on the key processing data of shipboard equipment at different time scales to obtain degradation analysis data at multiple time scales. The degradation analysis data at multiple time scales are compared with a preset time scale degradation threshold to obtain the time degradation comparison results; Based on the time degradation comparison results, degradation category determination is performed on degradation analysis data at multiple time scales to obtain degradation category determination information for time scale degradation analysis data; Degradation change analysis is performed based on degradation category determination information, and performance inflection point determination information is obtained based on degradation change analysis data.

[0009] Furthermore, the step of performing degradation change analysis based on degradation category determination information and obtaining performance inflection point determination information based on degradation change analysis data includes: Obtain the time-scale degradation analysis data difference for different degradation categories to obtain the degradation change difference; The degradation change difference is compared with a preset mutation threshold to obtain the degradation change comparison result; Based on the comparison results of the degradation changes, the performance inflection point of the time-scale degradation analysis data is determined to obtain performance inflection point determination information. The preset mutation threshold is set using a change point detection method.

[0010] Further, S4 includes: By using performance inflection point determination information, the entire life cycle of shipboard equipment is divided into segments to obtain multiple life segments. The lifespan segmentation includes a healthy stable period, a uniform degradation period, and an accelerated failure period; The model predictions of the preliminary life prediction model are compared with the actual monitoring values ​​to obtain output deviation comparison data. Based on the output deviation comparison data, the gradient descent method is used to obtain the segmented optimization factor. Based on the segmented optimization factor, the model parameters of multiple lifetime segments are adjusted to obtain model optimization data. Based on lifespan segmentation and segmented optimization factors, stage sub-models are obtained and adjusted and updated to obtain the final lifespan prediction model, thereby obtaining lifespan assessment strategy optimization data.

[0011] Furthermore, the step of obtaining stage sub-models based on lifetime segmentation and segmented optimization factors, adjusting and updating them to obtain the final lifetime prediction model, and then obtaining lifetime assessment strategy optimization data, includes: By combining multiple lifetime segments with corresponding segment optimization factors, the preliminary lifetime prediction model is divided into stages to obtain multiple stage sub-models. A stage-adjusted sub-model is obtained by adjusting the stage sub-model through segmented optimization factors. The stage sub-model is updated based on the node adjustment word model to obtain the updated sub-model; By connecting multiple stage sub-models and / or update sub-models, the final lifetime prediction model is obtained; Data for optimizing life assessment strategies is obtained through the final life prediction model.

[0012] Furthermore, the system includes: The data processing module is used to collect data from shipboard equipment, obtain shipboard equipment monitoring data, perform redundancy processing, information integration, compensation filling and expansion similarity analysis on the shipboard equipment monitoring data, and obtain key processing data of shipboard equipment. The model training module is used to train a deep learning model using key processing data from shipboard equipment, perform degradation feature training and analysis, and then build a preliminary life prediction model. The degradation analysis module is used to obtain time-scale degradation analysis data based on the preliminary life prediction model, perform degradation category analysis and determination based on the time-scale degradation analysis data, perform degradation change analysis based on the degradation category determination information, and obtain performance inflection point determination information based on the degradation change analysis data. The segmented optimization module is used to obtain multiple life segments of shipboard equipment based on performance inflection point determination information, perform output deviation analysis on the preliminary life prediction model, and then obtain segmented optimization factors. Based on the life segments and segmented optimization factors, stage sub-models are obtained and adjusted and updated to obtain the final life prediction model, and then life assessment strategy optimization data is obtained.

[0013] The beneficial effects of this invention are as follows: This invention achieves dynamic adaptation of the prediction model. Unlike traditional static models, the final model generated by this invention can identify changes in the equipment degradation stage and automatically switch to the corresponding prediction mode, thus maintaining high accuracy throughout the entire life cycle. By capturing performance inflection points, the system can issue early warnings when the equipment just enters the accelerated degradation phase, gaining valuable preparation time for maintenance decisions and realizing a shift from post-failure maintenance to pre-failure early warning. This invention is not a simple data fitting; through concepts such as health indices and performance inflection points, it incorporates the physical process of equipment degradation into the model, making the prediction results more accurate. The output of this invention includes not only the remaining lifespan but also optimization strategies based on lifespan segments, such as extending inspections during the healthy period and preparing spare parts during the accelerated degradation period. Attached Figure Description

[0014] Figure 1This is a schematic diagram of a deep learning-based optimization method for shipboard equipment life assessment strategies. Detailed Implementation

[0015] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0016] In one embodiment of the present invention, the present invention proposes a deep learning-based method and system for optimizing shipboard equipment life assessment strategies, the method comprising: S1. Collect shipboard equipment data to obtain shipboard equipment monitoring data, perform redundancy processing, information integration, compensation filling and expansion similarity analysis on the shipboard equipment monitoring data, and obtain key processing data of shipboard equipment. S2. Train a deep learning model using key processing data from shipboard equipment, perform degradation feature training and analysis, and then construct a preliminary life prediction model. S3. Obtain time-scale degradation analysis data based on the preliminary life prediction model, perform degradation category analysis and determination based on the time-scale degradation analysis data, perform degradation change analysis based on the degradation category determination information, and obtain performance inflection point determination information based on the degradation change analysis data. S4. Based on the performance inflection point determination information, obtain multiple life segments of shipboard equipment, perform output deviation analysis on the preliminary life prediction model, and then obtain segment optimization factors. Based on the life segments and the segment optimization factors, obtain stage sub-models and adjust and update them to obtain the final life prediction model, and then obtain life assessment strategy optimization data.

[0017] The working principle and technical effects of the above solution are as follows: Key information characterizing the health status of equipment is extracted from noisy raw data through data processing methods, particularly by converting the data into an image format that is easier for the model to understand. Deep learning is used to automatically learn degradation patterns from the images, establishing a preliminary prediction benchmark. The model incorporates historical data for time-series analysis, intelligently identifying the key performance inflection point from slow wear to accelerated aging. Based on this inflection point, the equipment lifespan is divided into different stages, and optimization strategies are tailored for each stage. The model parameters are dynamically adjusted through a feedback mechanism, ultimately forming a final prediction model that becomes increasingly accurate with use.

[0018] The training process of a deep learning model includes: This invention achieves dynamic adaptation of the prediction model. Unlike traditional static models, the final model generated by this invention can identify changes in the equipment degradation stage and automatically switch to the corresponding prediction mode, thus maintaining high accuracy throughout the entire lifecycle. By capturing performance inflection points, the system can issue early warnings when equipment just enters the accelerated degradation phase, gaining valuable preparation time for maintenance decisions and realizing a shift from post-failure maintenance to pre-failure early warning. This invention is not a simple data fitting; by incorporating concepts such as health indices and performance inflection points, it integrates the physical process of equipment degradation into the model, making the prediction results more accurate. The output of this invention includes not only the remaining lifespan but also optimization strategies based on lifespan segments, such as extending inspections during the healthy period and preparing spare parts during the accelerated degradation period.

[0019] Raw monitoring data of shipboard equipment is collected, and key processed data characterizing the health status of the equipment is extracted through redundancy removal, information integration, missing value compensation and filling, and similar data expansion analysis. The key processed data is converted into image form that the model can recognize, and input into a deep learning model for automatic extraction and training analysis of degradation features to build a preliminary life prediction model. Based on the time-scale degradation data output by the preliminary model, the degradation category of the equipment is determined and the degradation change pattern is analyzed to identify performance inflection points and divide the lifespan into segments. By analyzing the output deviation of the preliminary model, segment optimization factors are obtained, and stage sub-models are customized for each lifespan segment. After parameter adjustment and updating, the final life prediction model is formed.

[0020] In one embodiment of the present invention, S1 includes: Multi-source data collection is conducted on shipboard equipment to obtain multi-source data from shipboard equipment; Data collected from multiple shipboard equipment and sources is converted to a unified format to obtain multi-source converted data; the multi-source data is then uniformly organized through format conversion. Time-series data of multi-source converted data is obtained by time-labeling the collection time. Redundant data is removed from the collected time-series data to obtain redundant processed data; Redundant processed data is classified and stored across multiple collection categories to obtain a category-integrated database; Data missing information is obtained by identifying missing data in the category-integrated database. The missing information in the data is compensated and filled using time series prediction methods to obtain key processing data for shipboard equipment.

[0021] The working principle and technical effects of the above technical solution are as follows: Multi-source data acquisition ensures comprehensive information, such as simultaneously recording equipment vibration, temperature, and pressure. Standardized formatting converts all data into a standard language, eliminating communication barriers. Time stamping assigns precise timestamps to all data, establishing a sequence relationship of its changes over time. Redundancy removal removes duplicate and invalid information, reducing storage and computational burden. Categorized storage archives data by type (e.g., vibration database, temperature database) for easy management. Time series prediction (e.g., LSTM) intelligently fills in missing data gaps. The principle is to use the complete data context before and after the missing point to infer the most likely missing value, thus forming a complete, clean, and orderly set of critical processing data for shipboard equipment.

[0022] Through a series of rigorous preprocessing steps, the quality of input data is fundamentally improved, avoiding the dilemma of "garbage in, garbage out".

[0023] Using LSTM for data imputation is better at capturing temporal dependencies than traditional interpolation methods, and the imputation results are closer to the real physical process, significantly improving the utilization rate of incomplete datasets.

[0024] A unified standard for processing and analyzing data from different manufacturers and different types of sensors has been established, enhancing the system's versatility and scalability.

[0025] In one embodiment of the present invention, the step of compensating and filling in missing data information using a time series prediction method to obtain key processing data for shipboard equipment includes: Missing time-series data is obtained by collecting time-series data and combining it with time-missing information; By using long short-term memory networks to predict and fill in missing temporal data, integrated temporal data can be obtained. Spatial texture information is obtained by performing two-dimensional image transformation on temporally integrated data using the Gram angle field method; Spatial texture information is augmented and augmented similarity analysis is performed. Based on the similarity analysis data, key feature processing is carried out to obtain key processing data for shipboard equipment.

[0026] The working principle and technical effects of the above solution are as follows: Long Short-Term Memory (LSTM) networks are used to repair incomplete time-series data. Due to its unique memory gate structure, LSTM can effectively learn long-term dependencies, thus performing accurate filling. The repaired one-dimensional time series is converted into a two-dimensional image using the Gram Corner Field (GAF) method. Each time point data is mapped to a polar coordinate system, and the pixel values ​​of the image matrix are generated by calculating trigonometric functions between time points. This process cleverly encodes the temporal relationships and numerical correlations of the time series into the spatial structure and texture patterns of the image. A stable signal generates an image with uniform color gradients, while a signal containing periodic shocks generates an image with repeating textures. The spatial texture information contained in these images is precisely the visual fingerprint of the device degradation process.

[0027] By transforming a time-series problem into an image recognition problem, highly mature and powerful convolutional neural networks can be directly applied to automatically extract degenerative features. This is a technical shortcut that greatly improves the efficiency and effectiveness of feature extraction. Periodic and trend changes that are difficult to detect in one-dimensional data will appear as clear textures, patches, or lines in two-dimensional images, making the health status of devices visible and greatly enhancing the interpretability of the model. Image texture features are insensitive to random noise in the data, allowing the model to focus more on overall structural patterns, thereby improving the stability of predictions under complex conditions.

[0028] In one embodiment of the present invention, the process of augmenting spatial texture information and performing augmented similarity analysis, followed by key feature processing based on the similarity analysis data to obtain key processing data for shipboard equipment, includes: New augmentation points are generated between adjacent pixels of spatial texture information using bicubic interpolation to obtain an updated dataset; the updated dataset includes the original data and the augmentation point data. Obtain the similarity between the updated dataset and the corresponding data of adjacent pixels to obtain expanded similarity data; The expanded similarity data is compared with a preset similarity threshold to obtain a similarity comparison result; The updated dataset is assessed for eligibility based on the similarity comparison results to obtain similarity assessment information; Based on the similarity determination information, key feature data is extracted from the updated dataset to obtain key processing data for shipboard equipment. Similar data is extracted, while dissimilar data is expanded.

[0029] The working principle and technical effect of the above solution are as follows: New pixels (i.e., augmentation points) are calculated and inserted between the pixels in the image generated by GAF using bicubic interpolation. By considering a 4x4 pixel region, a very smooth and natural transition can be generated, making the image texture more delicate and conforming to the characteristics of continuous change in physical signals. The system calculates the similarity between the newly generated data (updated dataset) and the original data in terms of statistical distribution or features. A pass / fail judgment is made by comparing it with a preset similarity threshold: highly similar augmented data is retained for subsequent training; dissimilar data is considered defective, discarded, and regenerated to ensure the fidelity of the augmented data.

[0030] By augmenting the data with high-quality data, the training sample size of the deep learning model was significantly increased without conducting additional costly physical experiments, effectively preventing overfitting. A similarity quality control mechanism was introduced to ensure that all new data used for training the model strictly adhered to the physical laws of the original data, avoiding the introduction of false patterns that could lead to model misjudgments, thus improving the model's generalization ability and reliability. Images augmented through interpolation exhibit clearer and more continuous textures.

[0031] In one embodiment of the present invention, S2 includes: Deep learning models are trained using key data processed by shipboard equipment. Degradation feature data of key processing data of shipboard equipment is obtained through convolutional and pooling layers of deep learning models; A health index model is trained based on the degradation feature data, and health index trajectory data is output based on the health index model. Nonlinear regression analysis was performed on the health index trajectory data, and a preliminary lifespan prediction model was constructed based on the nonlinear regression analysis data.

[0032] The working principle and technical effects of the above solution are as follows: A deep learning model is trained using high-quality key processing data, with a Convolutional Neural Network (CNN) at its core. The convolutional layers of the CNN act like miniature feature filters, sliding across the image to automatically detect various local patterns such as edges, textures, and shapes. Pooling layers reduce the dimensionality of the extracted features, retaining the most significant information, improving computational efficiency, and enhancing robustness against interference. Through these layers, the model automatically learns and outputs degradation feature data directly related to device degradation. Based on these features, a health index model is trained, which maps a complex multidimensional feature vector to a single indicator (health index) between 0 and 1, representing the overall health of the device. Continuous health index values ​​constitute the health index trajectory. Finally, nonlinear regression analysis is performed on this trajectory to fit its function over time, thereby constructing a preliminary lifespan prediction model that can predict the remaining lifespan based on the current state.

[0033] Automatic and accurate extraction of degradation features has been achieved: It abandons the traditional manual feature extraction method that relies on expert experience, and utilizes CNNs to automatically learn optimal features from data (images), resulting in a more comprehensive and objective assessment. The health index condenses a complex multi-parameter degradation process into an intuitive numerical value, making equipment status assessment and comparison simple and consistent. The preliminary lifespan prediction model provides a reliable benchmark and starting point for model optimization and segmentation, giving the entire system a solid initial intelligence.

[0034] In one embodiment of the present invention, S3 includes: Based on the preliminary life prediction model and historical equipment operation data, degradation data analysis was performed on the key processing data of shipboard equipment at different time scales to obtain degradation analysis data at multiple time scales. The degradation analysis data at multiple time scales are compared with a preset time scale degradation threshold to obtain the time degradation comparison results; Based on the time degradation comparison results, degradation category determination is performed on degradation analysis data at multiple time scales to obtain degradation category determination information for time scale degradation analysis data; the degradation category determination information includes the time change pattern information of slow degradation period and accelerated degradation period.

[0035] Degradation change analysis is performed based on degradation category determination information, and performance inflection point determination information is obtained based on degradation change analysis data.

[0036] The working principle and technical effects of the above-mentioned technical solution are as follows: Based on a preliminary lifespan prediction model and combined with rich historical equipment operation data, a multi-angle, long-term retrospective analysis of the equipment status is conducted, namely, degradation data analysis at different time scales (e.g., analyzing monthly, quarterly, and annual degradation rates). By comparing the analysis results with time scale degradation thresholds preset based on extensive experience, the system can determine the current macroscopic degradation stage of the equipment, distinguishing between a slow degradation period and an accelerated degradation period, and summarizing its temporal variation patterns. Essentially, this creates a macroscopic degradation map for the equipment, identifying which macroscopic segment of the lifespan curve the equipment is currently in.

[0037] This allows for a macroscopic understanding of the overall degradation trend of equipment, determining whether it is in a stable, slow degradation phase or has entered a higher-risk, accelerated degradation phase. Through multi-timescale analysis, the model can filter out short-term fluctuations, focusing more on long-term, fundamental degradation patterns, thus improving the robustness of predictions. Accurate determination of the degradation category is a necessary preparation and foundation for the next step of precisely identifying the critical point (performance inflection point) between slow and accelerated degradation.

[0038] In one embodiment of the present invention, the step of performing degradation change analysis based on degradation category determination information and obtaining performance inflection point determination information based on degradation change analysis data includes: Obtain the time-scale degradation analysis data difference for different degradation categories to obtain the degradation change difference; The degradation change difference is compared with a preset mutation threshold to obtain the degradation change comparison result; Based on the comparison results of the degradation changes, the performance inflection point of the time-scale degradation analysis data is determined to obtain performance inflection point determination information. The preset mutation threshold is set using a change point detection method.

[0039] The working principle and technical effect of the above solution are as follows: By calculating the difference in degradation analysis data between adjacent time periods (e.g., from the slow phase to the accelerated phase), the intensity of the degradation rate change is quantified, i.e., the degradation change difference. This difference reflects whether the equipment is aging rapidly. The system then compares this difference with a key preset mutation threshold. The mutation threshold is usually learned from historical data through statistical methods such as change point detection, defining the boundary between normal rate change and abnormal mutation. If the degradation change difference exceeds this threshold, the system determines that there is a performance inflection point, indicating that the equipment degradation dynamics have undergone a qualitative change, moving from the accumulation stage of quantitative change to the accelerated stage of qualitative change.

[0040] The ability to automatically and accurately pinpoint inflection points on the equipment's lifespan curve is one of the most crucial and valuable pieces of information for predictive maintenance. Identifying performance inflection points provides clear and forward-looking timeframes for scheduling preventative maintenance and preparing spare parts, avoiding losses from unexpected failures. The existence of inflection points also provides objective and quantifiable dividing points for categorizing the entire equipment lifecycle into stages with different degradation characteristics.

[0041] In one embodiment of the present invention, S4 includes: By using performance inflection point determination information, the entire life cycle of shipboard equipment is divided into segments to obtain multiple life segments. The lifespan segmentation includes a healthy stable period, a uniform degradation period, and an accelerated failure period; The model predictions of the preliminary life prediction model are compared with the actual monitoring values ​​to obtain output deviation comparison data. Based on the output deviation comparison data, the gradient descent method is used to obtain the segmented optimization factor. Based on the segmented optimization factor, the model parameters of multiple lifetime segments are adjusted to obtain model optimization data. Based on lifespan segmentation and segmented optimization factors, stage sub-models are obtained and adjusted and updated to obtain the final lifespan prediction model, thereby obtaining lifespan assessment strategy optimization data.

[0042] The working principle and technical effects of the above solution are as follows: Utilizing the identified performance inflection points, the entire life cycle of the equipment is scientifically divided into several lifespan segments, including a healthy stable period, a uniform degradation period, and an accelerated failure period. The system compares the predicted values ​​of the preliminary model with the actual monitored values, generating output deviation comparison data. Using the gradient descent optimization algorithm, a segmented optimization factor is calculated for the data in different lifespan segments. This factor is essentially a calibration parameter used to correct the systematic bias of the preliminary model within that specific segment. By applying these factors, the model parameters are adjusted, making the model's predictions in each segment more closely match the actual situation.

[0043] Recognizing that the degradation mechanisms of equipment differ at different stages of its lifespan, this approach avoids a one-size-fits-all approach and achieves the highest predictive accuracy at each stage. Through feedback from output bias analysis and optimization factors, the model can continuously self-calibrate during use, becoming more accurate with each application and demonstrating a degree of self-learning capability. By tailoring the model to each stage, the final integrated model maintains excellent predictive performance throughout the entire equipment lifecycle, overcoming the drawback of traditional models that exhibit significant prediction bias in later stages.

[0044] In one embodiment of the present invention, the step of obtaining stage sub-models based on lifetime segmentation and segmentation optimization factors, adjusting and updating them to obtain a final lifetime prediction model, and then obtaining lifetime assessment strategy optimization data, includes: By combining multiple lifetime segments with corresponding segment optimization factors, the preliminary lifetime prediction model is divided into stages to obtain multiple stage sub-models. A stage-adjusted sub-model is obtained by adjusting the stage sub-model through segmented optimization factors. The stage sub-model is updated based on the node adjustment word model to obtain the updated sub-model; By connecting multiple stage sub-models and / or update sub-models, the final lifetime prediction model is obtained; Data for optimizing life assessment strategies is obtained through the final life prediction model.

[0045] The working principle and technical effects of the above solution are as follows: The pre-defined lifespan segments are combined with calculated segment optimization factors, breaking down the single initial lifespan prediction model into multiple stage-specific sub-models. Each sub-model is finely adjusted using its corresponding optimization factor, becoming a more specialized stage-adjusted or updated sub-model. These optimized sub-models are then connected using an algorithm (such as a smoothing transfer function) to form a unified, seamless final lifespan prediction model. This final model acts like a precision machine with multiple gears, automatically selecting the most suitable sub-model for prediction based on the current lifespan stage of the equipment, thus achieving accurate prediction from start to finish.

[0046] The final output is a stable and reliable ultimate solution capable of handling the entire complex degradation process of equipment. Through a clever model integration method, it ensures that the prediction results do not jump or drop abruptly when the equipment status spans different stages, resulting in a smooth and continuous output. This model outputs not just simple lifespan values, but lifespan assessment strategy optimization data that integrates stage information and can directly guide maintenance practices. For example, if the equipment has entered the accelerated phase, it is recommended to perform maintenance within three months, truly transforming the value of data analysis into decision-making value.

[0047] According to one embodiment of the present invention, the system includes: The data processing module is used to collect data from shipboard equipment, obtain shipboard equipment monitoring data, perform redundancy processing, information integration, compensation filling and expansion similarity analysis on the shipboard equipment monitoring data, and obtain key processing data of shipboard equipment. The model training module is used to train a deep learning model using key processing data from shipboard equipment, perform degradation feature training and analysis, and then build a preliminary life prediction model. The degradation analysis module is used to obtain time-scale degradation analysis data based on the preliminary life prediction model, perform degradation category analysis and determination based on the time-scale degradation analysis data, perform degradation change analysis based on the degradation category determination information, and obtain performance inflection point determination information based on the degradation change analysis data. The segmented optimization module is used to obtain multiple life segments of shipboard equipment based on performance inflection point determination information, perform output deviation analysis on the preliminary life prediction model, and then obtain segmented optimization factors. Based on the life segments and segmented optimization factors, stage sub-models are obtained and adjusted and updated to obtain the final life prediction model, and then life assessment strategy optimization data is obtained.

[0048] The working principle and technical effects of the above solution are as follows: Key information characterizing the health status of equipment is extracted from noisy raw data through data processing methods, particularly by converting the data into an image format that is easier for the model to understand. Deep learning is used to automatically learn degradation patterns from the images, establishing a preliminary prediction benchmark. The model incorporates historical data for time-series analysis, intelligently identifying the key performance inflection point from slow wear to accelerated aging. Based on this inflection point, the equipment lifespan is divided into different stages, and optimization strategies are tailored for each stage. The model parameters are dynamically adjusted through a feedback mechanism, ultimately forming a final prediction model that becomes increasingly accurate with use.

[0049] This invention achieves dynamic adaptation of the prediction model. Unlike traditional static models, the final model generated by this invention can identify changes in the equipment degradation stage and automatically switch to the corresponding prediction mode, thus maintaining high accuracy throughout the entire lifecycle. By capturing performance inflection points, the system can issue early warnings when equipment just enters the accelerated degradation phase, gaining valuable preparation time for maintenance decisions and realizing a shift from post-failure maintenance to pre-failure early warning. This invention is not a simple data fitting; by incorporating concepts such as health indices and performance inflection points, it integrates the physical process of equipment degradation into the model, making the prediction results more accurate. The output of this invention includes not only the remaining lifespan but also optimization strategies based on lifespan segments, such as extending inspections during the healthy period and preparing spare parts during the accelerated degradation period.

[0050] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for optimizing a strategy for evaluating the life of a marine equipment based on deep learning, characterized in that, The method comprises: S1, collecting ship equipment data to obtain ship equipment monitoring data, performing redundancy processing, information integration, compensation filling and expansion similarity analysis on the ship equipment monitoring data to obtain ship equipment key processing data; S2, training a deep learning model through the ship equipment key processing data, performing degradation feature training analysis, and then constructing a preliminary life prediction model; S3, obtaining time scale degradation analysis data according to the preliminary life prediction model, performing degradation category analysis and judgment according to the time scale degradation analysis data, performing degradation change analysis according to the degradation category judgment information, and obtaining performance inflection point judgment information according to the degradation change analysis data; S4, obtaining multiple life segments of the ship equipment according to the performance inflection point judgment information, performing output deviation analysis on the preliminary life prediction model, then obtaining a segmented optimization factor, obtaining a stage sub-model according to the life segment combined with the segmented optimization factor and adjusting and updating the stage sub-model to obtain a final life prediction model, and then obtaining life evaluation strategy optimization data.

2. The method of claim 1, wherein the method is characterized by, The S1 comprises: Collecting multi-source data of the ship equipment to obtain multi-source collected data of the ship equipment; Converting the multi-source collected data of the ship equipment into a uniform format to obtain multi-source converted data; Labeling the multi-source converted data with collection time to obtain collection time series data; Removing redundant data from the collection time series data to obtain redundancy processing data; Classifying and storing the redundancy processing data according to multiple collection categories to obtain a category integration database; Identifying data missing information from the category integration database; Compensating and filling the data missing information by a time series prediction method to obtain ship equipment key processing data.

3. The method of claim 2, wherein the method is characterized by, The method for compensating and filling the data missing information by a time series prediction method to obtain ship equipment key processing data comprises: Obtaining time sequence missing data by combining the collection time series data with the time missing information; Predicting and filling the time sequence missing data by a long short-term memory network to obtain time sequence integrated data; Converting the time sequence integrated data into a two-dimensional image by a Gram angle field method to obtain spatial texture information; Expanding and expanding similarity analyzing the spatial texture information, performing key feature processing according to the similarity analysis data, and obtaining ship equipment key processing data.

4. The method of claim 3, wherein the method is characterized by, The method for expanding and expanding similarity analyzing the spatial texture information, performing key feature processing according to the similarity analysis data, and obtaining ship equipment key processing data comprises: Generating new expansion points between adjacent pixel points of the spatial texture information by a bicubic interpolation method to obtain an updated data set; Obtaining similarity of the updated data set and corresponding data of adjacent pixel points to obtain expansion similarity data; Comparing the expansion similarity data with a preset similarity threshold to obtain a similarity comparison result; Performing qualified judgment on the updated data set according to the similarity comparison result to obtain similarity judgment information; Extracting key feature data of the updated data set according to the similarity judgment information to obtain ship equipment key processing data.

5. The method of claim 1, wherein the method is characterized by, The S2 comprises: Training a deep learning model through the ship equipment key processing data; Degeneration characteristic data of the key processing data of the marine equipment is obtained through a convolution layer and a pooling layer of a deep learning model; A health index model is trained according to the degeneration characteristic data, and health index trajectory data is output according to the health index model; Nonlinear regression analysis is performed on the health index trajectory data, and a preliminary life prediction model is constructed according to the nonlinear regression analysis data.

6. The method of claim 1, wherein the method is characterized by: The S3 comprises: Different time scale degeneration data analysis is performed on the key processing data of the marine equipment according to the preliminary life prediction model combined with historical equipment operation data, and a plurality of time scale degeneration analysis data is obtained; The plurality of time scale degeneration analysis data is compared with a preset time scale degeneration threshold, and a time degeneration comparison result is obtained; Degeneration category determination is performed on the plurality of time scale degeneration analysis data according to the time degeneration comparison result, and degeneration category determination information of the time scale degeneration analysis data is obtained; Degeneration change analysis is performed according to the degeneration category determination information, and performance inflection point determination information is obtained according to the degeneration change analysis data.

7. The method of claim 6, wherein the method is characterized by, The degeneration change analysis according to the degeneration category determination information and the performance inflection point determination information according to the degeneration change analysis data comprise: A time scale degeneration analysis data difference value of different degeneration category determination information is obtained, and a degeneration change difference value is obtained; The degeneration change difference value is compared with a preset mutation threshold, and a degeneration change comparison result is obtained; Performance inflection point determination is performed on the time scale degeneration analysis data according to the degeneration change comparison result, and performance inflection point determination information is obtained; The preset mutation threshold is set by a change point detection method.

8. The method of claim 1, wherein the method is characterized by: The S4 comprises: The marine equipment is segmented and divided through the performance inflection point determination information, and a plurality of life segments is obtained; The life segment comprises a health stable period, a uniform degeneration period and an accelerated failure period; The model prediction value of the preliminary life prediction model is compared with the actual monitoring value, and output deviation comparison data is obtained; A segmented optimization factor is obtained by using a gradient descent method according to the output deviation comparison data, the model parameter of the plurality of life segments is adjusted according to the segmented optimization factor, and model optimization data is obtained; A stage sub-model is obtained according to the life segment combined with the segmented optimization factor and is adjusted and updated, and a final life prediction model is obtained, and life evaluation strategy optimization data is further obtained.

9. The method of claim 8, wherein the method further comprises: The degeneration change analysis according to the degeneration category determination information and the performance inflection point determination information according to the degeneration change analysis data comprise: The plurality of life segments and the corresponding segmented optimization factor are combined, and the preliminary life prediction model is segmented, and a plurality of stage sub-models is obtained; The stage sub-model is adjusted through the segmented optimization factor, and a stage adjustment sub-model is obtained; The stage sub-model is updated according to the node adjustment sub-model, and an updated sub-model is obtained; The plurality of stage sub-models and / or updated sub-models are connected, and a final life prediction model is obtained; The life evaluation strategy optimization data is obtained through the final life prediction model.

10. A deep learning-based system for optimizing the life assessment strategy of a naval equipment, characterized by, The system comprises: The data processing module is configured to collect ship equipment data, obtain ship equipment monitoring data, perform redundancy processing, information integration, compensation filling, and expansion similarity analysis on the ship equipment monitoring data, and obtain key processing data of the ship equipment. The model training module is configured to train a deep learning model through the key processing data of the ship equipment, perform degradation feature training analysis, and further construct a preliminary life prediction model. The degradation analysis module is configured to obtain time scale degradation analysis data according to the preliminary life prediction model, perform degradation category analysis and determination according to the time scale degradation analysis data, perform degradation change analysis according to the degradation category determination information, and obtain performance inflection point determination information according to the degradation change analysis data. The segmented optimization module is configured to obtain a plurality of life segments of the ship equipment according to the performance inflection point determination information, perform output deviation analysis on the preliminary life prediction model, further obtain a segmented optimization factor, obtain a stage sub-model according to the life segments combined with the segmented optimization factor and perform adjustment and update, obtain a final life prediction model, and further obtain life evaluation strategy optimization data.

Images