Ship equipment maintenance support system and method based on artificial intelligence

Abstract

The invention discloses a ship equipment maintenance support system and method based on artificial intelligence, and the system collects the operation data of equipment in real time through a data collection and integration module, and digitalizes and integrates the historical maintenance data into a unified database; a fault diagnosis model is constructed through the model diagnosis module to carry out fault diagnosis and output a diagnosis result; an optimal maintenance scheme is provided through a maintenance decision support system according to a fault diagnosis result in combination with multi-dimensional data; real-time monitoring and trend analysis are carried out on operation data of the equipment through the fault prediction module, potential faults of the equipment are predicted, and a maintenance plan is made in advance. According to the invention, equipment operation data is collected in real time through a sensor, accurate fault diagnosis and prediction are realized in combination with a deep learning algorithm, an optimal maintenance scheme is formulated in combination with multi-dimensional data, and a maintenance plan is dynamically adjusted.

Description

Technical Field

[0001] The present invention relates to the field of ship equipment maintenance and support, and in particular to a ship equipment maintenance and support system and method based on artificial intelligence. Background Art

[0002] Currently, ship maintenance and support models rely primarily on traditional methods such as preventive maintenance, temporary emergency repairs, and empirical judgment. Maintenance personnel are required to regularly inspect equipment and systems, record their operating status, and perform repairs when equipment malfunctions.

[0003] Chinese patent publication number CN115809868A discloses "a ship maintenance method, system, terminal and storage medium". By obtaining historical information of various ship components, a degradation model of each ship component is obtained. Then, based on the status information of each ship component, it is determined whether each component is faulty. The status information of the fault-free components is introduced into the degradation model, and then the overall ship maintenance plan is obtained by combining the ship's itinerary information and the faulty component information.

[0004] However, the above scheme mainly relies on degradation models and status information for fault judgment. The data is relatively scattered, the utilization rate is low, and the diagnostic efficiency and accuracy are limited. Therefore, it is urgent to design an artificial intelligence-based ship equipment maintenance and support system and method to solve the problems existing in the above-mentioned existing technologies. Summary of the Invention

[0005] In response to the above-mentioned defects or improvement needs of the existing technology, the present invention provides a ship equipment maintenance and support system and method based on artificial intelligence, the purpose of which is to collect equipment operation data in real time through sensors, combine deep learning algorithms to achieve accurate fault diagnosis and prediction, combine multi-dimensional data to formulate the optimal maintenance plan, and dynamically adjust the maintenance plan.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions: A first aspect of the present invention provides an artificial intelligence-based ship equipment maintenance and support system, comprising: Data collection and integration module: used to collect equipment operation data in real time and digitally integrate historical maintenance data into a unified database to form a complete equipment operation and maintenance data set; Model diagnosis module: Use deep learning algorithms to model and train a complete equipment operation and maintenance data set, use the trained fault diagnosis model to perform fault diagnosis and output the diagnosis results; Maintenance Decision Support System: Provides the optimal maintenance plan based on fault diagnosis results, combined with multi-dimensional data on spare parts inventory, maintenance resources, and maintenance personnel skills; Fault prediction module: used to monitor and analyze the equipment's operating data in real time, predict potential equipment failures, and formulate maintenance plans in advance.

[0007] As an embodiment of the present application, the data collection and integration module includes: Data acquisition module: Multiple types of sensors are installed according to the classification of ship equipment to collect equipment operation data in real time through multiple types of sensors; Data digitization module: digitizes historical maintenance data in paper and electronic form and organizes it into a structured data format; Data integration module: used to integrate real-time collected operation data and historical maintenance data into a unified database to form a complete equipment operation and maintenance data set.

[0008] As an embodiment of the present application, the model diagnosis module includes: Data preprocessing module: used to preprocess the equipment's operating data and historical fault data, including data cleaning, feature extraction, and data labeling; Model training and optimization module: Use deep learning algorithms to model and train preprocessed data, build a fault diagnosis model, and optimize the model through a validation set to obtain a trained fault diagnosis model; Model deployment and application module: Deploy the trained fault diagnosis model to the maintenance decision support system, receive equipment operation data in real time, and output fault diagnosis results.

[0009] As an embodiment of the present application, the data cleaning includes outlier detection and processing, missing value repair and format unification; the feature extraction uses different methods to extract key features according to the data type; the data annotation includes fault label definition, expert calibration and imbalance processing.

[0010] As an embodiment of the present application, the system selects different deep learning algorithms to model and train the preprocessed data according to the data type, fault characteristics and data scale. The deep learning algorithms include convolutional neural networks, long short-term memory networks, Transformer networks and generative adversarial networks.

[0011] As an embodiment of the present application, the maintenance decision support system includes: Fault diagnosis result receiving module: used for receiving the fault diagnosis result of the fault diagnosis model; Multi-dimensional data integration module: used to integrate multi-dimensional data on inventory spare parts information, maintenance resources and maintenance personnel skills; Maintenance plan generation module: used to generate the optimal maintenance plan based on fault diagnosis results and multi-dimensional data.

[0012] As an embodiment of the present application, the optimal maintenance plan generates an optimal solution by quantitatively evaluating four indicators: maintenance time, maintenance cost, resource utilization, and maintenance quality. The comprehensive score of the four indicators is calculated using the hierarchical analysis method. The calculation formula for the comprehensive score is as follows:

[0013] in, represents the comprehensive score, Indicates maintenance time. represents the maintenance cost, Indicates resource utilization, Indicates the quality of repair.

[0014] As an embodiment of the present application, the fault prediction module includes: Operation data monitoring module: used to monitor the operation data of equipment collected by multiple types of sensors in real time; Trend analysis and prediction module: used to perform trend analysis on the equipment's operating data using time series analysis algorithms and predict potential equipment failures; Maintenance plan making module: used to make preventive maintenance plans in advance based on fault prediction results.

[0015] As an embodiment of the present application, the system uses different time series algorithms to perform trend analysis based on equipment data characteristics, failure modes, and scenario requirements. The time series analysis algorithms include autoregressive integral moving average model, seasonal autoregressive integral moving average model, long short-term memory network, and Prophet prediction model.

[0016] The present application also provides an artificial intelligence-based ship equipment maintenance and support method, which uses the above-mentioned artificial intelligence-based ship equipment maintenance and support system, and specifically includes the following steps: S1: Use multiple types of sensors to collect equipment operation data in real time, and digitally integrate historical maintenance data into a unified database to form a complete equipment operation and maintenance data set; S2: Use deep learning algorithms to model and train equipment operation and maintenance data sets, use the trained fault diagnosis model to perform fault diagnosis and output diagnosis results; S3: Based on the fault diagnosis results, combined with multi-dimensional data on spare parts inventory, maintenance resources, and maintenance personnel skills, the optimal maintenance plan is provided; S4: Conduct real-time monitoring and trend analysis of equipment operating data, predict potential equipment failures, and formulate maintenance plans in advance.

[0017] The beneficial effects of the present invention are: 1. The present invention preprocesses the equipment's operating data and historical fault data, models and trains the preprocessed data using a deep learning algorithm, constructs a fault diagnosis model, and optimizes the model using a validation set to obtain a trained fault diagnosis model. The trained fault diagnosis model is deployed into a maintenance decision support system, which receives the equipment's operating data in real time, extracts features from the data, performs precise analysis, quickly identifies the fault type and cause, and outputs a fault diagnosis result. Compared with traditional methods, the present invention significantly improves fault diagnosis accuracy and significantly shortens diagnosis time, thereby improving maintenance efficiency. 2. This invention uses a time series analysis algorithm to analyze the trends of equipment operating data, enabling early prediction of potential failures and the development of preventive maintenance plans. The failure prediction module effectively avoids unplanned downtime due to sudden equipment failures, reduces uncertainty in equipment operation, lowers equipment failure rates, and extends equipment life. 3. The present invention uses a maintenance decision support system to provide the optimal maintenance plan based on fault diagnosis results, combined with multi-dimensional data such as spare parts inventory information, maintenance resources, and maintenance personnel skills. This intelligent decision support model not only optimizes the maintenance process, but also improves resource utilization, avoiding excessive maintenance and resource waste. 4. The present invention uses multiple types of sensors to collect equipment operation data in real time and digitally integrates historical maintenance data to form a complete equipment operation and maintenance data set. This data integration model breaks the data island problem in the traditional maintenance model and provides a solid data foundation for the intelligent management of equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 This is a general architecture diagram of an artificial intelligence-based ship equipment maintenance and support system provided in an embodiment of the present invention; Figure 2 This is a flowchart of a data acquisition and integration module of an artificial intelligence-based ship equipment maintenance and support system provided in an embodiment of the present invention; Figure 3 This is a flowchart of a model diagnosis module of an artificial intelligence-based ship equipment maintenance and support system provided in an embodiment of the present invention; Figure 4 This is a flowchart of a maintenance decision support system for a ship equipment maintenance support system based on artificial intelligence provided in an embodiment of the present invention; Figure 5 This is a flowchart of a fault prediction module of an artificial intelligence-based ship equipment maintenance and support system provided in an embodiment of the present invention; Figure 6The present invention provides an artificial intelligence-based ship equipment maintenance and support method. DETAILED DESCRIPTION

[0019] 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. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0020] Reference Figure 1-Figure 5 The first aspect of the present invention provides a ship equipment maintenance and support system based on artificial intelligence, comprising: Data collection and integration module: used to collect equipment operation data in real time and digitally integrate historical maintenance data into a unified database to form a complete equipment operation and maintenance data set; Model diagnosis module: Use deep learning algorithms to model and train the equipment's operating data and historical fault data, use the trained fault diagnosis model to perform fault diagnosis and output the diagnosis results; Maintenance Decision Support System: Provides the optimal maintenance plan based on fault diagnosis results, combined with multi-dimensional data on spare parts inventory, maintenance resources, and maintenance personnel skills; Fault prediction module: used to monitor and analyze the equipment's operating data in real time, predict potential equipment failures, and formulate maintenance plans in advance.

[0021] The data acquisition and integration module in this application collects the operating data of the equipment in real time, such as temperature, vibration, pressure, etc., by installing sensors on the key equipment of the ship; at the same time, the historical maintenance data in paper and spreadsheet forms are digitized and integrated into a unified database to form a complete equipment operation and maintenance data set. This data integration model breaks the data silo problem in the traditional maintenance model and provides a solid data foundation for the intelligent management of equipment. In addition, the maintenance experience and knowledge are preserved in a digital form, avoiding the loss of experience due to personnel turnover. This knowledge inheritance model not only improves the overall skill level of the maintenance team, but also provides continuous support for the long-term operation and maintenance of the ship.

[0022] As an embodiment of the present application, the data collection and integration module includes: Data acquisition module: Multiple types of sensors are installed according to the classification of ship equipment to collect the equipment's operating data in real time. The installation configuration of multiple types of sensors is shown in Table 1:

[0023] Table 1 Installation configuration of multiple sensor types As can be seen from the table, multiple types of sensors are configured according to equipment characteristics, covering multi-dimensional parameters such as mechanical, electrical, and fluid. They can comprehensively monitor the operating status of ship equipment and promptly capture signals that may cause failures. This all-round monitoring is crucial for early problem identification and preventive maintenance.

[0024] Data digitization module: digitizes historical maintenance data in paper and spreadsheet formats and organizes them into a structured data format; the digitization process specifically includes data collection and entry, data cleaning and standardization, and structured storage. The collection and entry includes paper document processing and spreadsheet processing. The paper document processing converts paper work orders into images, such as PDF / A format, using a high-resolution scanner, and uses OCR (optical character recognition) technology to identify text content; the OCR technology recognition results are manually proofread to correct recognition errors, such as number confusion and loss of special symbols; the spreadsheet processing includes format unification and missing value completion. The format unification is used to standardize Excel spreadsheets from different departments or different periods according to a unified template to ensure field consistency; the missing value completion fills in the empty values ​​in the table by linking other data sources, such as supplementing the missing maintenance personnel ID in the table through equipment ledgers and maintenance logs.

[0025] Data cleaning and standardization include deduplication and error correction, as well as data standardization. Deduplication and error correction remove duplicate records and correct logical errors, such as removing multiple entries of the same maintenance event and correcting logical errors where the maintenance time is earlier than the fault occurrence time. Data standardization includes standardizing time stamps, establishing a standardized fault coding system for fault classification, such as "E001 - Bearing Wear" and "E002 - Impeller Crack," mapping text descriptions to codes, and standardizing physical quantity units. Structured storage stores data in standardized database tables, such as MySQL and PostgreSQL. Free text in maintenance reports is stored in JSON format to facilitate subsequent text mining.

[0026] Data integration module: used to integrate real-time collected operation data and historical maintenance data into a unified database. Specifically, it includes data synchronization architecture design, data model design, time alignment and association, consistency verification and quality control, and finally forms a complete equipment operation and maintenance data set.

[0027] The synchronization architecture design includes the extraction, conversion, and loading of real-time and historical data. First, the sensor data stream is received through the message queue to extract real-time data; incremental data is extracted from the relational database or data lake at regular intervals to extract historical data; real-time data is aggregated through windows, such as calculating the average and maximum values ​​once a minute; real-time data is formatted, such as converting the maintenance timestamp to a time zone consistent with the sensor data; the processed real-time and historical data are written into a unified data warehouse or time series database.

[0028] The data model design adopts a star architecture, and the core realizes efficient analysis and query of data through the association between fact tables and dimension tables. The fact table stores equipment operation indicators and maintenance time, and the dimension table contains static data such as equipment information, fault classification, and maintenance personnel.

[0029] The time alignment and association accurately associates device events with sensor data in the time dimension through time window matching and causal relationship marking to support scenarios such as fault analysis and status monitoring; the time window matching associates the maintenance time with the sensor data window before the fault occurs, and the causal relationship marking clarifies the causal relationship between the device status and the maintenance event by adding a maintenance time mark in the real-time data stream. For example, after the fault is repaired, the subsequent data is marked as normal operation.

[0030] The consistency verification and quality control include data integrity check and conflict resolution decision. The data integrity check verifies whether the device ID and timestamp of real-time data match those of historical data, and detects abnormal values ​​based on the physical characteristics of the device, such as sensor readings exceeding the physical limits of the device. When real-time data conflicts with historical records, the conflict resolution decision gives priority to real-time data and triggers manual review.

[0031] The model diagnosis module in this application uses a deep learning algorithm to analyze and model the equipment's operating data and historical fault data to construct a fault diagnosis model; it extracts features from the equipment's operating data and performs precise analysis to quickly identify the type and cause of the fault. Compared with traditional methods, the fault diagnosis accuracy of the present invention is significantly improved, the diagnosis time is greatly shortened, manual intervention is reduced, and the accuracy of diagnosis is improved.

[0032] As an embodiment of the present application, the model diagnosis module includes: Data preprocessing module: used to preprocess the equipment's operating data and historical fault data. The preprocessing includes data cleaning, feature extraction, and data labeling.

[0033] Specifically, the data cleaning includes outlier detection and processing, missing value repair and format unification; in the outlier detection and processing, the 3σ principle or IQR (interquartile range) method is used to detect outliers. The 3σ principle assumes that the data obeys the normal distribution, calculates the mean and standard deviation, and regards data values ​​that exceed the range of mean ± 3 times the standard deviation as outliers. The IQR (interquartile range) method defines the abnormal range by the interquartile range, and usually regards data values ​​that exceed Q1-1.5IQR or Q3+1.5IQR as outliers; for single-point outliers, adjacent valid value interpolation is used, such as the previous moment and the next moment The average value of the data at each moment is combined with the equipment operation logic judgment for continuous abnormal segments. For example, the speed of a diesel engine should be 0 when it is shut down. If a non-zero value is collected, it is marked as invalid or filled with historical data of similar equipment. The missing value repair method uses linear interpolation or cubic spline interpolation to fill short-term missing data (such as less than five minutes), and uses historical data of similar equipment operating conditions to fill long-term missing data. The format unification includes timestamp alignment and unit standardization. The timestamp alignment synchronizes the collected data of different sensors according to millisecond timestamps and aggregates them through time windows, such as taking the average / maximum value per second. The unit is standardized to a unified physical quantity unit, such as temperature is unified as ℃, vibration amplitude is unified as g, and pressure is unified as MPa. After the above processing, the data cleaning can remove noise, errors and inconsistencies in the data, ensuring the accuracy and reliability of the input data.

[0034] The feature extraction method uses different methods to extract key features based on the data type. Specifically, it includes: extracting statistical features, trend features, and frequency domain features from time series data (such as temperature, pressure, and speed). Statistical features include mean, variance, peak, root mean square (RMS), kurtosis, and skewness, reflecting the data distribution characteristics. Trend features use a sliding window to calculate slopes, such as the temperature change rate per minute, to identify upward or downward trends. Frequency domain features perform a Fourier transform on time series data to extract the primary and secondary frequencies and their amplitudes, which is suitable for periodic signals such as gear mesh frequency. Time-frequency analysis and envelope analysis are used to capture transient faults in vibration signals (non-stationary signals). Time domain analysis uses a short-time Fourier transform (STFT) to generate a time-frequency plot or a wavelet transform to extract the time-frequency energy distribution, capturing transient impact faults such as bearing cracks. Envelope analysis performs a Hilbert transform on vibration signals to extract fault characteristic frequencies, such as those of bearing inner and outer race faults. Multi-sensor fusion features are extracted, concatenating temperature, vibration, and pressure features from the same device into a multidimensional feature vector (e.g., [mean temperature, root mean square vibration, peak pressure]), or a tensor is constructed and fed into a deep learning model. This process allows feature extraction to extract the key information that best reflects the device's status from the raw data, reducing data dimensionality and enhancing feature representation.

[0035] Data labeling involves fault label definition, expert calibration, and imbalance processing. Specifically, fault label definition integrates historical maintenance records, such as work orders and fault reports, to associate data segments with fault type, severity, and occurrence time. For example, if a diesel engine vibration signal exhibits an anomaly at a certain point in time, corresponding to the maintenance record "mild bearing wear," the data for the 30 minutes before and after that period will be labeled "bearing wear - mild." Labeling granularity is based on time windows (e.g., 1 minute), and faults within the window containing the fault onset are assigned the corresponding fault category. Expert calibration is performed for fuzzy data, such as those with partial anomalies in multi-sensor signals, and manual labeling by experienced marine engineers ensures label accuracy. Imbalance processing addresses situations with a small number of fault samples, such as "impeller fracture" with only five records. SMOTE oversampling or the ADASYN algorithm is used to generate synthetic samples. Alternatively, a loss function, such as Focus Loss, can be used to assign higher weight to the minority class. After this process, data labeling assigns fault labels to data samples, enabling the model to learn the mapping between fault modes and equipment status.

[0036] Model training and optimization module: Use deep learning algorithms to model and train preprocessed data, build a fault diagnosis model, and optimize the model through a validation set to obtain a trained fault diagnosis model; Specifically, the system selects different deep learning algorithms to model and train the pre-processed data according to the data type, fault characteristics and data scale. The deep learning algorithms include convolutional neural networks (CNN), long short-term memory networks (LSTM), transformer networks and generative adversarial networks (GAN). The data types include: vibration signals, select convolutional neural networks (CNN); time series data, select long short-term memory networks (LSTM) or transformer networks; multimodal data, select transformer networks; the fault characteristics include: periodic faults, select long short-term memory networks (LSTM) to capture time series characteristics; instantaneous impact faults, select convolutional neural networks (CNN) to extract local features. The data scale includes: for small sample scenarios, selecting Generative Adversarial Network (GAN) to enhance data; for large-scale data, selecting Convolutional Neural Network (CNN) or Long Short-Term Memory Network (LSTM); specifically, the Convolutional Neural Network (CNN) is suitable for data with spatial characteristics such as vibration signals and images, such as the time-frequency diagram of gearbox vibration signals; the Long Short-Term Memory Network (LSTM) is suitable for time series data, such as the temporal changes of diesel engine speed and temperature; the Transformer network is suitable for multi-sensor data fusion scenarios to capture the long-term dependencies between different signals; the Generative Adversarial Network (GAN) is used to generate abnormal samples to solve the problem of imbalanced fault data, and a small number of fault samples are enhanced by Generative Adversarial Network (GAN). The above algorithms can be flexibly combined in predictive maintenance of ship equipment according to data characteristics, such as linearity or nonlinearity, seasonality, and business needs, such as short-term anomaly detection or long-term trend analysis.

[0037] The fault diagnosis model training and optimization process includes using different loss functions for different task categories, cross entropy for classification problems, and mean square error for regression problems; then using the Adam optimizer to automatically adjust the learning rate of each parameter, and finally verifying and optimizing through 5-fold cross-validation and early stopping strategy to obtain a trained fault diagnosis model.

[0038] Model deployment and application module: Deploy the trained fault diagnosis model to the maintenance decision support system, receive equipment operation data in real time, and output fault diagnosis results.

[0039] Specifically, the fault diagnosis model diagnostic process includes real-time data input, feature extraction, fault classification and decision threshold; the real-time data input is real-time data collected by multiple types of sensors, sliced ​​according to time windows (such as 10 seconds), and converted into a model input format with the dimensions of [number of samples, time steps, number of features]; the feature extraction automatically extracts deep features through a convolutional neural network (CNN) or a long short-term memory network (LSTM) layer, such as the fault feature frequency of the vibration signal and the inflection point of the temperature trend; the fault classification outputs the probability of each fault category by the fully connected layer, such as "bearing wear: 95%, normal: 3%, other: 2%"; the decision threshold is a set probability threshold, such as a probability ≥90% to trigger a fault alarm, combined with expert rules, such as "vibration amplitude > threshold and temperature abnormality" to determine it as a bearing fault.

[0040] The maintenance decision support system in this application provides the optimal maintenance plan based on fault diagnosis results, combined with multi-dimensional data such as spare parts inventory, maintenance resources, and maintenance personnel skills. This intelligent decision support model not only optimizes the maintenance process but also improves resource utilization, avoiding excessive maintenance and resource waste.

[0041] As an embodiment of the present application, the maintenance decision support system includes: Fault diagnosis result receiving module: used for receiving the fault diagnosis result of the fault diagnosis model; Multidimensional Data Integration Module: This module integrates multidimensional data on inventory spare parts, maintenance resources, and maintenance personnel skills. Specifically, this module integrates multidimensional data on inventory spare parts, maintenance resources, and maintenance personnel skills through data modeling and standardization, data extraction and integration architecture, relationship building, data quality control and real-time synchronization, and a data service layer. This data modeling and standardization eliminates data heterogeneity across different systems by defining globally unique identifiers (such as equipment IDs, spare part codes, and personnel IDs), developing a unified data dictionary (e.g., fault classification codes and maintenance priorities), and establishing metadata standards.

[0042] The data extraction and integration architecture extracts data through the API of the spare parts management system, the database of the personnel management system, and file import. It cleans data by processing negative numbers in the spare parts inventory and invalid certificates of personnel skills, standardizes the timestamps of different systems to unify the format, maps spare parts categories such as "bearings" to a unified coding system for data mapping, and then completes data conversion; the data after data conversion is loaded into a data warehouse or NoSQL database, and Apache Kafka is used to process real-time data streams, such as spare parts outbound notifications. Airflow is used to orchestrate scheduled ETL tasks, such as daily synchronization of personnel skills updates, ultimately achieving real-time or scheduled data acquisition and conversion from heterogeneous data sources.

[0043] The association relationship is established by defining the BOM association between spare parts and equipment and the mapping relationship between personnel skills and tasks, and using a graph database (such as Neo4j) to store complex relationships, support fast path queries, and ultimately build logical associations between data, thereby supporting cross-dimensional data analysis and decision-making in complex business scenarios.

[0044] The data quality control and real-time synchronization are implemented by verifying the consistency of spare parts inventory and purchase orders, performing integrity checks, and checking the validity period of personnel skill certificates to achieve logical verification. The message queue is used to trigger real-time updates of spare parts inventory changes and synchronize personnel skill changes through the HR system API to ensure the accuracy, completeness and timeliness of the data.

[0045] The data service layer provides efficient support for upper-layer applications through a unified data access interface, and uses Redis to cache high-frequency query data, such as common spare parts inventory, significantly improving data access response speed and system performance.

[0046] Maintenance plan generation module: used to generate the optimal maintenance plan based on fault diagnosis results and multi-dimensional data.

[0047] As an embodiment of the present application, the optimal maintenance plan generates an optimal solution by quantitatively evaluating four indicators: maintenance time, maintenance cost, resource utilization, and maintenance quality. The comprehensive score of the four indicators is calculated using the Analytic Hierarchy Process (AHP). The calculation formula for the comprehensive score is as follows:

[0048] in, represents the comprehensive score, Indicates maintenance time. represents the maintenance cost, Indicates resource utilization, Indicates the quality of repair.

[0049] Specifically, the analytic hierarchy process (AHP) normalizes the four indicators of maintenance time, maintenance cost, resource utilization, and maintenance quality, and performs weighted calculation using preset weights. The solution with the highest comprehensive score is the optimal solution.

[0050] Among them, the weight of maintenance time is set at 40%, which affects the number of ship stoppages. Priority is given to the shortest solution. The length of maintenance time is directly related to the ship's sailing efficiency and operating costs, so time is one of the most important considerations; the weight of maintenance cost is set at 30%, including spare parts costs, labor costs, downtime losses, etc. Maintenance cost is a direct reflection of economic efficiency and needs to be minimized while ensuring maintenance quality. The weight of resource utilization is set at 20%, including spare parts inventory turnover rate, maintenance personnel idle rate, etc. Resource utilization reflects the rational allocation and efficiency of resources during the maintenance process; the weight of maintenance quality is set at 10%, which refers to the historical success rate of the maintenance team, such as the port team success rate of 95% and the onboard self-repair success rate of 85%. Maintenance quality is a key factor in ensuring the long-term stable operation of equipment.

[0051] Through this systematic evaluation method, it is possible to ensure that in complex maintenance scenarios, the influence of various factors is comprehensively considered and the best maintenance plan that suits the current situation is selected, thereby improving maintenance efficiency and resource utilization and reducing maintenance costs. In addition, through the long-term accumulation and analysis of equipment operation data and maintenance data, support can be provided for the full life cycle management of the equipment. From equipment design, manufacturing, operation to retirement, the present invention can provide data-driven optimization suggestions, thereby improving the efficiency and economy of equipment use.

[0052] For example, if the diagnosis results indicate that a ship's gearbox has experienced gear wear, the steps of the multi-dimensional data integration module are as follows: Fault diagnosis result: Gear wear (severity: moderate), repair recommended within 48 hours; Inventory spare parts information: The inventory quantity of gear spare parts is 2 pieces, and the nearest port warehouse can allocate 3 pieces. The arrival time is 12 hours. Maintenance resources: The current skill level of the ship's maintenance personnel is that 3 personnel are qualified to disassemble and assemble the gearbox, and 1 person is qualified to perform advanced diagnostics.

[0053] Cost constraint: port maintenance labor cost (2,000 yuan / hour) vs. onboard repair (500 yuan / hour, but requires 2 crew members).

[0054] Generate a plan: Option 1: Self-study on board, takes 8 hours Resource utilization: Consume inventory spare parts, improve turnover rate, use existing personnel, no idleness, resource utilization rate is 80% Maintenance quality: mainly mid-level personnel, with a historical success rate of 85% for similar maintenance Cost: 8 x 500 = 4,000 yuan, 8 hours Option 2: Port maintenance, maintenance after docking at the port, takes 6 hours Resource utilization: Spare parts need to be transported across warehouses, resulting in a low turnover rate. The port team may have to wait for dispatch. The port professional team + spare parts dispatch, the arrival time is 12 hours, and the total time is 12+6=18 hours, with a resource utilization rate of 60%. Repair quality: Professional team's historical success rate for similar repairs is 95% Cost: 6 x 2000 = 12000 yuan, 18 hours The comprehensive scores are shown in Table 2:

[0055] Table 2 Comprehensive score As can be seen from Table 2, due to the maintenance time in the formula and maintenance costs With a total weight of 70%, Option 1 has a significant advantage due to its shorter time and lower cost, and has the highest comprehensive score, making it the optimal maintenance option. Although Option 2 has higher maintenance quality, the long suspension of service and high cost result in a lower comprehensive score, making it more suitable for scenarios with extremely high requirements for maintenance quality and insensitivity to time costs.

[0056] The present invention combines the experience of maintenance personnel with artificial intelligence technology through human-machine collaboration to form a new maintenance model. This method not only improves maintenance efficiency, but also can give full play to the subjective initiative of maintenance personnel, providing more flexible support for ship maintenance.

[0057] The fault prediction module in this application uses algorithms such as time series analysis to conduct real-time monitoring and trend analysis of equipment operation data, predict potential equipment failures, formulate maintenance plans in advance, avoid unplanned shutdowns caused by equipment failures, increase the ship's sailing rate, reduce uncertainty in equipment operation, reduce equipment failure rate, and extend the service life of the equipment.

[0058] As an embodiment of the present application, the fault prediction module includes: Operation data monitoring module: used to monitor the operation data of equipment collected by multiple types of sensors in real time; Trend analysis and prediction module: used to perform trend analysis on the equipment's operating data using time series analysis algorithms and predict potential equipment failures; Maintenance plan making module: used to make preventive maintenance plans in advance based on fault prediction results.

[0059] As an embodiment of the present application, the system uses different time series analysis algorithms to perform trend analysis based on equipment data characteristics, failure modes, and scenario requirements. The time series analysis algorithms include autoregressive integrated moving average model (ARIMA), seasonal autoregressive integrated moving average model (SARIMA), long short-term memory network (LSTM), and Prophet prediction model. Among them, the autoregressive integrated moving average model (ARIMA) is suitable for linear trend data, such as the temperature of a stably operating motor; the seasonal autoregressive integrated moving average model (SARIMA) takes seasonality into account, such as equipment with periodic load changes; the long short-term memory network (LSTM) captures nonlinear trends, such as the gradual wear of diesel engine vibration signals; the Prophet prediction model handles abnormal points such as holidays / working condition switching, and is suitable for equipment with complex working conditions.

[0060] Specifically, the time series analysis algorithm first reduces noise and normalizes the time series data, then extracts trend terms through polynomial fitting, Fourier transforms periodic terms, and detects anomalies through Z-score. Hourly data from 30 days of history is input for training, and predicted values ​​for the next 24 hours are output. Fault prediction is performed by calculating the deviation between the predicted value and the threshold (e.g., the temperature prediction value exceeds the alarm threshold + 5°C for 3 consecutive hours) or by combining it with a historical fault case library (e.g., "continuous overtemperature + increased vibration amplitude" corresponds to bearing lubrication failure). A corresponding maintenance plan is formulated according to the fault level. When a minor fault occurs, maintenance is planned at the next port. When a moderate fault occurs, onboard self-repair is planned within 48 hours. When a severe fault occurs, an emergency plan is triggered and the unit is immediately shut down for maintenance. The fault level maintenance plan table is shown in Table 3:

[0061] Table 3 Fault level maintenance plan For example, ARIMA predicts that the flow data of a pump equipment on a ship will be less than 80% of the rated value in the next 24 hours, showing a continuous downward trend. Combined with the pressure sensor data, the outlet pressure fluctuation increases. The system determines that this is the early stage of "impeller wear" and generates a plan to replace the impeller three days in advance to avoid sudden downtime.

[0062] Through precise fault diagnosis and preventive maintenance, this application can proactively prepare the necessary personnel, equipment, materials, and other elements for maintenance, reducing unplanned downtime and maintaining the ship's performance. This not only increases the ship's sailing rate but also enhances its operational effectiveness in complex mission environments. On the one hand, it reduces the cost of repairs due to sudden equipment failures; on the other hand, it avoids excessive maintenance and resource waste, and improves the utilization rate of maintenance resources. This not only reduces the ship's operating and maintenance costs, but also improves its economic benefits, generating positive ecological benefits for the sustainable development of ships.

[0063] like Figure 6 As shown, the present application also provides an artificial intelligence-based ship equipment maintenance and support method, which uses the above-mentioned artificial intelligence-based ship equipment maintenance and support system, specifically including the following steps: S1: Use multiple types of sensors to collect equipment operation data in real time, and digitally integrate historical maintenance data into a unified database to form a complete equipment operation and maintenance data set; S2: Use deep learning algorithms to model and train equipment operation and maintenance data sets, use the trained fault diagnosis model to perform fault diagnosis and output diagnosis results; S3: Based on the fault diagnosis results, combined with multi-dimensional data on spare parts inventory, maintenance resources, and maintenance personnel skills, the optimal maintenance plan is provided; S4: Conduct real-time monitoring and trend analysis of equipment operating data, predict potential equipment failures, and formulate maintenance plans in advance.

[0064] The above description is merely an illustration of some preferred embodiments of the present disclosure and the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present disclosure is not limited to the technical solutions formed by the specific combination of the above-mentioned technical features, but also includes other technical solutions formed by any combination of the above-mentioned technical features or their equivalents without departing from the above-mentioned inventive concept.

Claims

1. A ship equipment maintenance and support system based on artificial intelligence, characterized in that: include: Data collection and integration module: used to collect equipment operation data in real time and digitally integrate historical maintenance data into a unified database to form a complete equipment operation and maintenance data set; Model diagnosis module: Use deep learning algorithms to model and train a complete equipment operation and maintenance data set, use the trained fault diagnosis model to perform fault diagnosis and output the diagnosis results; Maintenance Decision Support System: Provides the optimal maintenance plan based on fault diagnosis results, combined with multi-dimensional data on spare parts inventory, maintenance resources, and maintenance personnel skills; Fault prediction module: used to monitor and analyze the equipment's operating data in real time, predict potential equipment failures, and formulate maintenance plans in advance.

2. The artificial intelligence-based ship equipment maintenance support system according to claim 1, characterized in that: The data collection and integration module includes: Data acquisition module: Multiple types of sensors are installed according to the classification of ship equipment to collect equipment operation data in real time through multiple types of sensors; Data digitization module: digitizes historical maintenance data in paper and electronic form and organizes it into a structured data format; Data integration module: used to integrate real-time collected operation data and digitized historical maintenance data into a unified database to form a complete equipment operation and maintenance data set.

3. The artificial intelligence-based ship equipment maintenance support system according to claim 1, characterized in that: The model diagnosis module includes: Data preprocessing module: used to preprocess the equipment's operating data and historical fault data, including data cleaning, feature extraction, and data labeling; Model training and optimization module: Use deep learning algorithms to model and train preprocessed data, build a fault diagnosis model, and optimize the model through a validation set to obtain a trained fault diagnosis model; Model deployment and application module: Deploy the trained fault diagnosis model to the maintenance decision support system, receive equipment operation data in real time, and output fault diagnosis results.

4. The artificial intelligence-based ship equipment maintenance support system according to claim 3 is characterized in that: The data cleaning includes outlier detection and processing, missing value repair and format unification; the feature extraction uses different methods to extract key features according to the data type; and the data annotation includes fault label definition, expert calibration and imbalance processing.

5. The artificial intelligence-based ship equipment maintenance support system according to claim 3 is characterized in that: The system selects different deep learning algorithms to model and train the preprocessed data according to the data type, fault characteristics and data scale. The deep learning algorithms include convolutional neural networks, long short-term memory networks, transformer networks and generative adversarial networks.

6. The artificial intelligence-based ship equipment maintenance support system according to claim 1, characterized in that: The maintenance decision support system includes: Fault diagnosis result receiving module: used for receiving the fault diagnosis result of the fault diagnosis model; Multi-dimensional data integration module: used to integrate multi-dimensional data on inventory spare parts information, maintenance resources and maintenance personnel skills; Maintenance plan generation module: used to generate the optimal maintenance plan based on fault diagnosis results and multi-dimensional data.

7. The artificial intelligence-based ship equipment maintenance support system according to claim 6, characterized in that: The optimal maintenance plan is generated by quantitatively evaluating the four indicators of maintenance time, maintenance cost, resource utilization, and maintenance quality. The comprehensive score of the four indicators is calculated using the hierarchical analysis method. The calculation formula of the comprehensive score is as follows: in, represents the comprehensive score, Indicates maintenance time. represents the maintenance cost, Indicates resource utilization, Indicates the quality of repair.

8. The artificial intelligence-based ship equipment maintenance support system according to claim 1, characterized in that: The fault prediction module includes: Operation data monitoring module: used to monitor the operation data of equipment collected by multiple types of sensors in real time; Trend analysis and prediction module: used to perform trend analysis on the equipment's operating data using time series analysis algorithms and predict potential equipment failures; Maintenance plan making module: used to make preventive maintenance plans in advance based on fault prediction results.

9. The artificial intelligence-based ship equipment maintenance support system according to claim 8, characterized in that: The system uses different time series analysis algorithms to perform trend analysis based on equipment data characteristics, failure modes, and scenario requirements. The time series analysis algorithms include autoregressive integrated moving average model, seasonal autoregressive integrated moving average model, long short-term memory network, and Prophet prediction model.

10. A ship equipment maintenance and support method based on artificial intelligence, characterized in that: The artificial intelligence-based ship equipment maintenance support system according to any one of claims 1 to 9 specifically comprises the following steps: S1: Use multiple types of sensors to collect equipment operation data in real time, and digitally integrate historical maintenance data into a unified database to form a complete equipment operation and maintenance data set; S2: Use deep learning algorithms to model and train equipment operation and maintenance data sets, use the trained fault diagnosis model to perform fault diagnosis and output diagnosis results; S3: Based on the fault diagnosis results, combined with multi-dimensional data on spare parts inventory, maintenance resources, and maintenance personnel skills, the optimal maintenance plan is provided; S4: Conduct real-time monitoring and trend analysis of equipment operating data, predict potential equipment failures, and formulate maintenance plans in advance.

Images