Marine engine operation and maintenance method, system and terminal based on vibration signal and LLM

Abstract

The invention provides a marine engine operation and maintenance method, system and terminal based on vibration signals and LLM, and the method comprises the steps: firstly collecting original vibration signals and crankshaft angle signals of each ship engine vibration measurement point, preprocessing the original vibration signals, carrying out the statistics of typical features, and storing related signals in a recent preset time period; secondly, identifying abnormal measuring points based on typical features, and constructing a parameter data set for the abnormal measuring points; and then, inputting the data set into a vertical field large language model, calling different tools for analysis, and retrieving and outputting a fault analysis report by using a knowledge database. And when operation and maintenance personnel inquire about the analysis result, the model also retrieves and answers through the knowledge database. Cognitive reasoning of a large language model (LLM) is combined with tool calling capacity of a Function Calling technology, automatic diagnosis and natural language interaction are achieved, the method is deployed on specific optimization equipment, most problems can be solved without expert support under the scene that an ocean vessel network is poor and communication cost is high, operation and maintenance cost is reduced, and the method is low in use threshold and easy to implement. And an operation and maintenance user interacts with the large language model to obtain various operation and maintenance information and details.

Description

Technical Field

[0001] This invention relates to the field of marine engine condition monitoring technology, and in particular to a marine engine operation and maintenance method, system and terminal based on vibration signals and LLM. Background Technology

[0002] In the critical field of industrial equipment condition monitoring, vibration signal analysis has always been regarded as a core technical means to predict equipment failures and ensure long-term stable and reliable operation. This is particularly crucial for marine engine condition monitoring. Traditional marine engine condition monitoring methods heavily rely on expert experience during the diagnostic process. Experts analyze and judge vibration signals based on years of accumulated practical experience and professional knowledge. However, this reliance on expert experience leads to low diagnostic efficiency. Each analysis of marine engine vibration signals requires a significant investment of time and effort from experts, often taking several hours. During ship navigation, equipment failures can occur at any time, and such lengthy diagnostic times can easily miss the optimal maintenance window, leading to further deterioration of the fault, increased maintenance costs, and equipment downtime. Furthermore, the fragmented nature of expert experience makes knowledge transfer extremely difficult. The experience accumulated by experts over long periods of work largely exists in personal memories and notes, lacking systematic organization and summarization. Newly hired maintenance personnel cannot fully grasp this experience in a short time and require extensive practical exploration and oral instruction from experts. Furthermore, the valuable experience of senior experts who leave or retire may be lost, resulting in significant losses for marine engine condition monitoring. In addition, traditional methods are inadequate in terms of dynamic adaptability. Marine engines face various complex operating conditions during actual operation, and accurate diagnosis of equipment failures is impossible when these conditions change.

[0003] Especially in the unique environment of ocean-going vessels, the network environment is poor and communication costs are high. When marine engines malfunction, remote support from experts is often required for diagnosis and repair guidance. However, due to network limitations, experts cannot obtain detailed operating data and vibration signals of the equipment in real time, affecting the accuracy and timeliness of remote diagnosis. Moreover, obtaining expert support requires paying high communication and expert service fees, further increasing operation and maintenance costs. At the same time, due to poor communication and limited expert support, maintenance efficiency is also very low, failing to meet the requirements of ocean-going vessels for rapid equipment repair and efficient operation. Summary of the Invention

[0004] In view of the above-mentioned disadvantages of the prior art, the purpose of the present application is to provide a marine engine operation and maintenance method and system based on vibration signals and LLM, and a terminal, to solve the technical problems of low diagnosis efficiency, difficult knowledge inheritance, poor dynamic adaptability of the existing marine engine state monitoring technology relying on expert experience, insufficient accuracy and timeliness of remote diagnosis under the scene of ocean-going ships due to network limitations, high operation and maintenance cost, and low maintenance efficiency.

[0005] To achieve the above-mentioned purpose and other related purposes, the present application provides a marine engine operation and maintenance method based on vibration signals and LLM, which comprises: collecting original vibration signals and crank angle signals of each marine engine vibration measuring point; preprocessing the original vibration signals of each marine engine vibration measuring point and counting typical features, and simultaneously storing the original vibration signals and crank angle signals of each marine engine vibration measuring point in the latest preset time period; identifying the abnormality of each marine engine vibration measuring point based on the counted typical features, and constructing a parameter data set for the marine engine vibration measuring point identified as abnormal; inputting each parameter data set into a vertical field large language model, calling different tools for analysis and using a knowledge database for retrieval to output a fault analysis report, and when natural language interaction information input by an operation and maintenance personnel in the vertical field large language model is asked for the fault analysis report, the knowledge database is used for retrieval and corresponding answer information is output.

[0006] In an embodiment of the present application, the sampling frequency and sampling time of the original vibration signals and crank angle signals of each marine engine vibration measuring point are determined according to the engine speed range and cycle period in combination with the vibration characteristics of each marine engine vibration measuring point.

[0007] In an embodiment of the present application, each marine engine vibration measuring point includes: an engine cylinder head, an engine cylinder liner, and an engine frame; wherein the original vibration signals of each marine engine vibration measuring point are collected by vibration sensors respectively arranged on the engine cylinder head, the engine cylinder liner, and the engine frame, and the crank angle signals are collected by a crank angle collecting device arranged at the engine flywheel disc.

[0008] In an embodiment of the present application, the preprocessing of the original vibration signals of each marine engine vibration measuring point and counting the typical features includes: filtering and trend item removal of the original vibration signals of each marine engine vibration measuring point; time domain feature and frequency domain feature extraction of the vibration signals after filtering and trend item removal, to obtain the typical features of each marine engine vibration measuring point and store them.

[0009] In an embodiment of the present application, the statistical-based typical features are used to identify the abnormality of each ship engine vibration measuring point, and the parameter data set of the ship engine vibration measuring point identified as abnormal is constructed by: inputting the typical features of each ship engine vibration measuring point into the constructed abnormality detection model to identify the abnormality, and obtaining the abnormality identification result of each ship engine vibration measuring point; locking the abnormal ship engine vibration measuring point based on the abnormality identification result, and integrating the original vibration signal, the typical features, the crank angle signal and the external system parameters of the corresponding ship engine vibration measuring point in the latest preset time period to obtain the parameter data set of each ship engine vibration measuring point.

[0010] In an embodiment of the present application, the parameter data set is input into the vertical field large language model, different tools are called for analysis, and the knowledge database is used for retrieval to output the fault analysis report, which includes: sequentially calling the adaptive tools in the tool set to analyze the input parameter data set in the vertical field large language model to obtain the preliminary analysis result, then retrieving the related knowledge from the knowledge database according to the preliminary analysis result, then calling the adaptive tools from the tool set again to analyze the retrieved knowledge, and continuing to retrieve in the knowledge database based on the new analysis result, through cyclic iteration, until the perfect fault analysis result is obtained; generating the fault analysis report based on the fault analysis result and drawing the visual chart to output.

[0011] In an embodiment of the present application, the tool set includes: analysis tools configured based on each ship engine vibration measuring point, and simulation tools configured based on each fault scene; the vertical field large language model can call the corresponding analysis tools for analysis according to the data belonging to the measuring point, and when the data is detected to conform to a fault scene, the corresponding simulation tool is called for analysis.

[0012] In an embodiment of the present application, the information fed back by the operation and maintenance personnel for the fault analysis report and the answer information is used to optimize the vertical field large language model.

[0013] To achieve the above object and other related objects, the present application provides a marine engine operation and maintenance system based on vibration signals and LLM, which comprises: a data acquisition module for acquiring original vibration signals and crank angle signals of each marine engine vibration measuring point; a feature statistics module connected to the data acquisition module for preprocessing the original vibration signals of each marine engine vibration measuring point and counting typical features; a storage module connected to the feature statistics module and the data acquisition module, comprising: a time series database and a memory database, respectively used for long-term storage of the counted typical features and short-term storage of the original vibration signals and crank angle signals of each marine engine vibration measuring point within a preset time period; an abnormality identification module connected to the storage module for identifying abnormality of each marine engine vibration measuring point based on the counted typical features, and constructing a parameter data set for the marine engine vibration measuring point identified as abnormal; and a vertical domain large language model analysis and interaction module connected to the abnormality identification module for inputting each parameter data set into the vertical domain large language model, calling different tools for analysis and using a knowledge database for retrieval to output a fault analysis report, and when natural language interaction information input by an operation and maintenance personnel is input into the vertical domain large language model for inquiry on the fault analysis result, the knowledge database is used for retrieval to output corresponding answer information.

[0014] To achieve the above object and other related objects, the present application provides an electronic terminal, comprising: one or more memories and one or more processors; the one or more memories are used for storing computer programs; the one or more processors are connected to the memories and are used for running the computer programs to execute the marine engine operation and maintenance method based on vibration signals and LLM.

[0015] As described above, the present application is a marine engine operation and maintenance method, system and terminal based on vibration signals and LLM, which has the following beneficial effects: the present application first acquires original vibration signals and crank angle signals of each marine engine vibration measuring point, preprocesses the original vibration signals and counts typical features, and simultaneously stores related signals within a preset time period. Then, the typical features are used to identify abnormal measuring points, and a parameter data set is constructed for the abnormal measuring points. After that, the data set is input into the vertical domain large language model, different tools are called for analysis, and a knowledge database is used for retrieval to output a fault analysis report. When an operation and maintenance personnel inquires about the analysis result, the model also answers through the knowledge database. The present application combines the cognitive reasoning of the large language model (LLM) with the tool calling ability of the Function Calling technology, realizes automatic diagnosis and natural language interaction, and is deployed on a specific optimized device. In the scene of poor network and high communication cost of ocean-going ships, most problems can be solved without expert support, the operation and maintenance cost is reduced, the threshold for use is low, and the operation and maintenance user can obtain various operation and maintenance information and details by interacting with the large language model. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 A flowchart of a marine engine operation and maintenance method based on vibration signals and LLM according to an embodiment of the present application is shown.

[0017] Figure 2 A vibration measurement point diagram of a marine engine according to an embodiment of the present application is shown.

[0018] Figure 3 A flowchart of a marine engine operation and maintenance method based on vibration signals and LLM according to an embodiment of the present application is shown.

[0019] Figure 4 A structure diagram of a marine engine operation and maintenance system based on vibration signals and LLM according to an embodiment of the present application is shown.

[0020] Figure 5 A structure diagram of an electronic terminal according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0021] The present application is herein described, by way of example only, with reference to the accompanying drawings, FIG. 1 illustrates a marine engine operation and maintenance system based on vibration signals and LLM according to an embodiment of the present application. FIG. 2 illustrates a marine engine vibration measurement point diagram according to an embodiment of the present application. FIG. 3 illustrates a flowchart of a marine engine operation and maintenance method based on vibration signals and LLM according to an embodiment of the present application. FIG. 4 illustrates a structure diagram of a marine engine operation and maintenance system based on vibration signals and LLM according to an embodiment of the present application. FIG. 5 illustrates a structure diagram of an electronic terminal according to an embodiment of the present application. Other embodiments of the present application will be apparent to those skilled in the art from consideration of the specification and practice of the application disclosed herein. Various substitutions and modifications can be made by those skilled in the art, without departing from the spirit or scope of the application as defined by the accompanying claims. It is intended that the specification and examples be considered as exemplary only, with the true scope and spirit of the application being indicated only by the following claims.

[0022] It is to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting. It is also possible in the present disclosure that steps can be executed in different sequence where is practical unless otherwise specified or explicitly denied herein. Furthermore, various

[0023] Throughout the specification, when it is said that a certain part is "connected" to another part, this includes not only the case of "direct connection" but also the case of "indirect connection" in which other elements are interposed therebetween. In addition, when it is said that a certain part "includes" a certain constituent element, other constituent elements are not excluded unless specifically stated to the contrary, and it means that other constituent elements can be further included.

[0024] The terms first, second, third, etc. mentioned herein are used to describe various parts, components, regions, layers and / or sections, but are not limited thereto. These terms are used only to distinguish a certain part, component, region, layer or section from another part, component, region, layer or section. Therefore, the first part, component, region, layer or section described below can be referred to as the second part, component, region, layer or section within the scope of the present application.

[0025] Furthermore, as used herein, the singular forms "a", "an" and "the" are intended to include plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises", "comprising", "includes" and / or "including" mean that the described features, operations, elements, components, items, species, and / or groups are present, but do not preclude the presence or addition of one or more other features, operations, elements, components, items, species, and / or groups. The terms "or" and "and / or" used herein are interpreted as inclusive, or mean either one or any combination. Thus, "A, B or C" or "A, B and / or C" means "any of the following: A; B; C; A and B; A and C; B and C; A, B and C". This definition applies only when a combination of elements, functions or operations are inherently mutually exclusive.

[0026] The present application provides a marine engine operation and maintenance method based on vibration signals and LLM. First, the original vibration signals and crank angle signals of each marine engine vibration measuring point are collected. The original vibration signals are preprocessed and the typical features are counted, and the related signals in the recent preset time period are stored. Then, based on the typical feature identification of the abnormal measuring point, a parameter data set is constructed for the abnormal measuring point. Then, the data set is input into the vertical field large language model, different tools are called for analysis, and the knowledge database is searched to output a fault analysis report. When the operation and maintenance personnel ask about the analysis results, the model also answers through the knowledge database search. The present application combines the cognitive reasoning of the large language model (LLM) with the tool calling ability of the Function Calling technology, realizes automatic diagnosis and natural language interaction, and is deployed on a specific optimized device. In the scene of poor network and high communication cost of ocean-going ships, most problems can be solved without expert support, reducing operation and maintenance costs, and the use threshold is low. The operation and maintenance user interacts with the large language model to obtain various operation and maintenance information and details.

[0027] With reference to the drawings, the embodiments of the present application will be described in detail below, so that those skilled in the art of the technical field to which the present application pertains can easily implement the present application. The present application can be embodied in various different forms and is not limited to the embodiments described herein.

[0028] As Figure 1 A flowchart of a marine engine operation and maintenance method based on vibration signals and LLM in an embodiment of the present application is shown.

[0029] The method comprises:

[0030] Step S1: Collecting original vibration signals of each ship engine vibration measuring point and crank angle signals.

[0031] In an embodiment, the selection of the vibration measuring point needs to determine the vibration signal characteristics, analysis requirements, and intelligent technology and other factors. Based on this, the ship engine vibration measuring point focuses on the engine cylinder head, cylinder sleeve, and frame three key components, and synchronously analyzes in combination with the crank angle signal.

[0032] As Figure 2 , by arranging vibration sensors on the cylinder head, cylinder sleeve, and frame, the original vibration signals of each measuring point are directly collected; a crank angle acquisition device (Hall sensor or rotary encoder) is arranged at the engine flywheel disc to collect the crank angle signal, which is used for synchronous analysis in combination with the engine operating cycle (such as correlation cycle characteristics and processing of non-stationary signals).

[0033] For the ship engine vibration measuring point as the engine cylinder head, it is usually measured synchronously with the crank angle of the crankshaft driving end or free end, which can detect valve failure and abnormal combustion, and needs to be synchronized with the engine cycle to process non-stationary signals; for the ship engine vibration measuring point as the cylinder sleeve, it focuses on high-frequency components and statistical indicators, which is suitable for detecting wear and can improve the automation level of diagnosis with the help of machine learning; for the ship engine vibration measuring point as the frame, it can evaluate the overall vibration severity and identify the dominant frequency at the main body position of the frame, and the order tracking is suitable for variable speed working conditions, and can also optimize the installation system; in the area close to the crankshaft and bearing of the frame, it can detect and analyze vibration abnormalities caused by crankshaft imbalance, bearing failure or bearing bush wear.

[0034] In an embodiment, in the actual ship engine vibration signal collection work, the sampling frequency and sampling time of the original vibration signals of each ship engine vibration measuring point and the crank angle signals need to be accurately determined, and this process is closely related to the engine speed range, cycle period, and vibration characteristics of each measuring point.

[0035] Firstly, the metadata of the acquisition channel, such as sensor ID or channel label, can directly and explicitly indicate the position source of the signal, providing a basis for subsequent targeted processing. Then, the engine speed range and cycle period are analyzed in depth. At different speeds, the vibration of each component of the engine is significantly different, and the cycle period reflects the repeated rules of engine operation. At the same time, the vibration characteristics of each ship engine vibration measuring point are fully considered, such as the cylinder head measuring point for detecting valve failure and abnormal combustion, the vibration signal of which has a specific frequency range and variation law; the cylinder sleeve measuring point focuses on high-frequency components to detect wear, and the vibration characteristics are different from other parts; the vibration characteristics of the machine frame measuring point are also complex and diverse when evaluating the overall vibration or detecting specific faults. Considering the above factors, the appropriate sampling frequency and sampling time of each measuring point are determined, and then the continuous sampling work is started to ensure that the collected signals are complete and accurate, providing a reliable basis for subsequent fault diagnosis and analysis.

[0036] Step S2: preprocessing the original vibration signal of each ship engine vibration measuring point and counting the typical features, and at the same time storing the original vibration signal and the crank angle signal of each ship engine vibration measuring point in the latest preset time period.

[0037] In an embodiment, the preprocessing of the original vibration signal of each ship engine vibration measuring point and counting the typical features comprises:

[0038] As Figure 3 , first, for the original vibration signal collected by each ship engine vibration measuring point, filtering and trend item removal operations are implemented. Filtering can effectively eliminate noise interference in the signal, remove those unrelated to the true vibration state of the device, and make the signal more pure; and trend item removal can eliminate linear or nonlinear trends in the signal caused by environmental factors, sensor drift, etc., so that the signal better reflects the vibration characteristics of the device itself. After filtering and trend item removal, the processed vibration signal is feature extracted. On the one hand, statistical features that can represent abnormal characteristics are extracted in the time domain, these statistical features include mean, variance, peak value and other indicators, which can reflect the signal changes in the time domain from different angles, providing an important basis for judging whether the device is abnormal. On the other hand, the processed signal is subjected to fast Fourier transform to obtain the frequency spectrum, and then frequency domain statistical features such as amplitude and power of each frequency component are extracted, which helps to analyze the frequency distribution of the device vibration and identify the frequency characteristics corresponding to specific faults. Finally, the typical features of each ship engine vibration measuring point extracted are properly stored for subsequent analysis and use.

[0039] In an embodiment, while carrying out preprocessing work on the original vibration signals of each ship machine vibration measuring point, in order to improve the real-time performance and availability of the data, the original vibration signals of each ship machine vibration measuring point in the last preset time period and the crank angle signals are stored. Specifically, the original vibration signals are cached to the memory database Redis, because in actual application, the number of vibration channels is large, and the collection frequency is high, which leads to extremely large amount of data. In view of this, it is not appropriate to directly store the original signals for a long time. However, in the field of vibration analysis, a large number of analysis methods depend on the original vibration signals. Therefore, the last 3 to 5 seconds of original vibration signals are stored in Redis, in order to further analyze the signals near the abnormal point when an abnormality is detected. This operation can ensure that when recent original vibration data is needed, there is no need to obtain complete signals, greatly shortening the data acquisition time, and providing strong support for subsequent possible real-time analysis and fault judgment. At the same time, in order to comprehensively record the running state of the ship machine and ensure the integrity of data analysis, the collected crank angle signals also need to be stored. The crank angle signal is closely related to the vibration signal, and it can reflect the running period and phase information of the engine. Combined with the stored crank angle signal, the correlation between the vibration signal and the running state of the engine can be more accurately analyzed, so that the equipment fault can be more effectively identified, and reliable guarantee can be provided for the stable operation of the ship machine.

[0040] Step S3: Abnormalities of each ship machine vibration measuring point are identified based on statistical typical features, and a parameter data set is constructed for the ship machine vibration measuring points identified as abnormal.

[0041] In an embodiment, step S3 includes:

[0042] As Figure 3 , first, the typical features obtained after preprocessing and feature extraction of each ship machine vibration measuring point are input into the pre-constructed abnormality detection model. Because fault data is almost impossible to obtain, and because of the particularity of the ship machine, it is undoubtedly not worth the loss to conduct a fault injection experiment on it. This abnormality detection model can use a single classification algorithm to wrap the data distribution of normal data, and has the ability to accurately analyze and judge the vibration features. With the help of this model, normal ship machine vibration measuring points can be quickly and accurately identified, and it can be determined which measuring points have abnormal conditions.

[0043] After obtaining the abnormal identification result, the ship machine vibration measuring point where the abnormality occurs is accurately locked according to the result. Then, comprehensive data integration is performed on these abnormal measuring points. The integrated data sources are extensive, including not only the cached original vibration signals of the abnormal measuring point in the recent preset time period (such as the last 3s-5s), which can directly reflect the recent vibration state of the measuring point, but also the stored crank angle signals, which are closely related to the engine operation cycle and are helpful for in-depth analysis of the cause of vibration abnormality. At the same time, the previously counted typical features are also included in the integration range, and external system parameters such as engine temperature, speed, load, etc. are introduced, which can provide more comprehensive background information for fault analysis. By integrating these multi-dimensional data, the parameter data set of each ship machine vibration measuring point is finally obtained, providing a solid data foundation for subsequent in-depth analysis and processing of faults.

[0044] Step S4: inputting each parameter data set into the vertical field large language model, calling different tools for analysis and using the knowledge database for retrieval to output a fault analysis report, and when natural language interaction information is input into the vertical field large language model for the fault analysis result inquiry of the operation and maintenance personnel, the knowledge database is used for retrieval and corresponding answer information is output.

[0045] In an embodiment, the inputting each parameter data set into the vertical field large language model, calling different tools for analysis and using the knowledge database for retrieval to output a fault analysis report comprises:

[0046] Firstly, the vertical field large language model can accurately judge which advanced analysis tools need to be used according to the ship machine vibration measuring point information in the input parameter data set. With the help of Function Calling technology, the adaptive tools are called in the preset tool set to analyze each parameter data set in the model, and then the preliminary analysis result is obtained.

[0047] Subsequently, according to the preliminary analysis result, the model retrieves related knowledge from the knowledge database. The knowledge database is a knowledge graph constructed based on the ISO fault feature library and historical data, which is like a huge fault knowledge base, and can effectively match signal features (such as specific frequency peaks) with fault modes (such as bearing wear and misalignment), providing accurate basis for fault analysis. Through retrieval, the model can obtain potential fault information and solutions related to the current analysis result.

[0048] If the retrieved knowledge is not sufficient, the vertical domain large language model will again select appropriate tools from the tool set, further analyze the retrieved knowledge, and continue searching in the knowledge database based on the new analysis results. This process forms a looped iterative thinking chain, just like when interacting with a large model, the model can determine its next operation direction based on the hints or knowledge we provide. The system continuously digs into the association between data and knowledge, constantly enriches and perfects the analysis results.

[0049] During the entire analysis process, whether to continue to call tools depends on the system's thinking amount, that is, the system determines whether to further call tools to obtain more accurate results based on the current analysis situation. Through such a looped iterative way, until the perfect fault analysis result is obtained.

[0050] To improve model performance and deployment efficiency, the vertical domain large language model uses devices packaged by the NVIDIA Jetson series module for quantization optimization and deployment. The model can fully utilize the signal source information and pre-training knowledge to directly select appropriate analysis methods, further improving the accuracy and efficiency of analysis, and providing strong support for precise diagnosis of ship machinery vibration faults.

[0051] Finally, the vertical domain large language model integrates the perfect fault analysis results to generate a detailed fault analysis report. The report content covers conclusions, basis, and suggestions, and calls the drawing function to generate time domain, frequency domain, or order graph visualization charts for output, which intuitively displays signal characteristics, making it easy for operation and maintenance personnel to quickly understand the fault situation and take appropriate measures.

[0052] In an embodiment, the tool set includes:

[0053] Analysis tools configured based on each ship machinery vibration measuring point: Since the physical quantities, signal characteristics, and roles played in device operation of different ship machinery vibration measuring points are different, a dedicated analysis tool is customized for each measuring point.

[0054] For the engine cylinder head measuring point, the monitored vibration signal contains rich information closely related to combustion conditions, valve working states, etc. For this measuring point, event synchronization analysis and time-frequency analysis are mainly used, and the corresponding analysis tools are called.

[0055] For the engine cylinder sleeve measuring point, the monitored vibration signal mainly reflects the friction, impact between the piston and the cylinder sleeve, and the vibration characteristics of the cylinder sleeve itself. For this measuring point, frequency spectrum analysis, statistical feature extraction, and machine learning are mainly used, and the corresponding analysis tools are called.

[0056] For engine frame measurement points, the monitored vibration signals contain comprehensive information of the coupling of various engine components. For this measurement point, overall vibration monitoring, frequency spectrum analysis, and order tracking are mainly used, and corresponding analysis tools are called.

[0057] Simulation tools configured for each fault scenario: During operation, the ship engine may encounter various fault scenarios, such as bearing wear, gear tooth breakage, piston ring breakage, etc. For each fault scenario, a corresponding simulation tool is configured.

[0058] During analysis, the vertical domain large language model monitors whether the data conforms to the characteristics of a known fault scenario in real time. Once it detects that the data matches a certain fault scenario, the model will immediately call the corresponding simulation tool for analysis. For example, if the analysis finds that the vibration signal of a certain measurement point contains frequency components consistent with the characteristics of gear tooth breakage, the model will call the simulation tool for the gear tooth breakage fault scenario to further simulate the vibration response under this fault and compare it with the actual data to verify whether the fault has occurred and its severity.

[0059] In an embodiment, the ship engine generates a large amount of complex vibration data during operation, and the vertical domain large language model analyzes these data in depth and generates a fault analysis report. However, the fault analysis report is rich in content and highly professional, and the operation and maintenance personnel may have difficulty quickly finding the specific details they are interested in. To more accurately and efficiently obtain the required information, the operation and maintenance personnel interact with the large language model through natural language, asking specific questions about the fault analysis report, and the model can quickly locate and return accurate answers.

[0060] The operation and maintenance personnel input questions in the form of natural language in the interactive interface, and the questions are closely related to the specific details in the fault analysis report. For example, after viewing the engine cylinder head related fault analysis, input "What is the abnormal frequency of the cylinder head signal?" or "What time point does the cylinder head fault appear, and is it recorded in which part of the report?" etc. This natural language input method conforms to the daily communication habits of the operation and maintenance personnel, and does not require the mastery of complex query instructions or programming languages.

[0061] After receiving the question input by the operation and maintenance personnel, the vertical domain large language model uses natural language processing techniques to analyze the question. This includes word segmentation, part-of-speech tagging, named entity recognition, etc. to accurately understand the key information in the question.

[0062] In generating the fault analysis report, the vertical domain large language model establishes detailed indexes for each part of the report. These indexes cover chapter titles, paragraph contents, key data, and other information in the report, and are associated with specific locations in the report. The model performs a quick search in the index of the knowledge database based on the key information obtained after analyzing the problem. The model extracts the key information related to the problem from the searched report location. If the problem involves multiple aspects of information or the searched information is scattered, the model integrates the information. The model generates a natural and fluent answer using natural language generation technology based on the extracted and integrated information. The answer is expressed in clear and understandable language, which conforms to the communication habits of the operation and maintenance personnel. For example, for the question "What is the abnormal frequency of the cylinder cover signal?", the model may answer "The abnormal frequency of the cylinder cover signal is between 1500-1800 Hz." Then, the answer is output to the interactive interface for the operation and maintenance personnel to view.

[0063] In an embodiment, in the ship engine operation and maintenance scenario, after the vertical domain large language model (LLM) provides the fault analysis report and answer for the operation and maintenance personnel, the operation and maintenance personnel will give feedback from the aspects of diagnosis accuracy and suggestion feasibility. The feedback is input through a special entry of the interactive interface. Based on this, the LLM adopts targeted optimization strategies: on the one hand, the tool calling logic is optimized, and the selection and use of diagnosis tools and maintenance suggestion generation tools are adjusted according to the feedback; on the other hand, the model parameters are optimized through supervised learning fine-tuning and reinforcement learning, and the feedback is used as new labeled data or reward signal to improve the model's diagnosis accuracy and suggestion feasibility. Facing new working conditions and fault modes, the LLM updates the knowledge graph and diagnosis model using incremental learning. Incremental learning can integrate new fault data and correct results into the training set without retraining the entire model, add new entity relationships, adjust knowledge weights, and adjust the model structure according to the characteristics of new faults, so that the model adapts to changes. At the same time, the optimization effect is evaluated through indicators such as diagnosis accuracy, suggestion adoption rate, and user satisfaction. The diagnosis accuracy reflects the model's fault recognition ability, the suggestion adoption rate reflects the practicality of the suggestion, and the user satisfaction understands whether the model meets the needs, so as to realize continuous improvement.

[0064] Similar to the principles of the above embodiments, the present application provides a ship engine operation and maintenance method based on vibration signals and LLM.

[0065] The following provides specific embodiments in combination with the drawings:

[0066] As Figure 4 A structure diagram of a ship engine operation and maintenance system based on vibration signals and LLM in an embodiment of the present application is shown.

[0067] The system comprises:

[0068] A data acquisition module 1 is configured to acquire original vibration signals of each marine engine vibration measuring point and a crank angle signal;

[0069] A feature statistics module 2 is connected to the data acquisition module 1 and configured to pre-process the original vibration signals of each marine engine vibration measuring point and to count typical features.

[0070] A storage module 3 is connected to the feature statistics module 2 and the data acquisition module 1 and includes a time series database and a memory database. The time series database is responsible for long-term and ordered storage of the typical features processed by the feature statistics module 3, so as to ensure data persistence and traceability. The memory database is configured to efficiently and quickly store the original vibration signals of each marine engine vibration measuring point and the crank angle signal in a preset time period, so as to meet the requirements of real-time monitoring and instant analysis. An abnormality identification module 4 is connected to the storage module 3 and configured to identify abnormalities of each marine engine vibration measuring point based on the counted typical features and to construct a parameter data set for the marine engine vibration measuring point identified as abnormal.

[0071] A vertical domain large language model analysis and interaction module 5 is connected to the abnormality identification module 4 and configured to input each parameter data set into a vertical domain large language model, to call different tools for analysis and to use a knowledge database for retrieval to output a fault analysis report. When natural language interaction information input by an operation and maintenance personnel in the vertical domain large language model is input, the knowledge database is used for retrieval and corresponding answer information is output.

[0072] Since the implementation principle of the marine engine operation and maintenance system based on vibration signals and LLM has been described in the foregoing embodiments, it will not be repeated here.

[0073] The marine engine operation and maintenance method based on vibration signals and LLM provided by the embodiment of the present application can be implemented on the terminal side or the server side. As for the hardware structure of the electronic terminal, please refer to Figure 5 , an optional hardware structure schematic diagram of the electronic terminal 1000 provided by the embodiment of the present application. The terminal 1000 can be a mobile phone, a computer device, a tablet device, a personal digital processing device, a factory background processing device, etc. The terminal 1000 includes at least one processor 1001, a memory 1002, at least one network interface 10010 and a user interface 1009. Each component in the device is coupled together through a bus system 1005. It can be understood that the bus system 1005 is used to realize the connection communication between the components. The bus system 1005 includes a data bus, a power bus, a control bus and a state signal bus. However, for the purpose of clear illustration, all kinds of buses are marked as a bus system in Figure 5 .

[0074] The user interface 1009 can include a display, a keyboard, a mouse, a trackball, a pointing gun, a key, a button, a touchpad, a touch screen, or the like.

[0075] It can be understood that the memory 1002 can be a volatile memory or a non-volatile memory, and can also include both volatile and non-volatile memories. The non-volatile memory can be a read-only memory (ROM), a programmable read-only memory (PROM), and used as an external cache. By way of example but not limitation, many forms of RAM can be used, such as static random access memory (SRAM), synchronous static random access memory (SSRAM). The memory described in the embodiments of the present application is intended to include but not limited to these and any other suitable categories of memory.

[0076] The memory 1002 in the embodiments of the present application is used to store various categories of data to support the operation of the terminal 1000. Examples of these data include: any executable programs for operating on the terminal 1000, such as an operating system 10021 and an application program 10022; the operating system 10021 contains various system programs, such as a framework layer, a core library layer, a driver layer, etc., for implementing various basic services and processing hardware-based tasks. The application program 10022 can contain various application programs, such as a media player (MediaPlayer), a browser (Browser), etc., for implementing various application services. The implementation of the vibration signal and LLM-based marine engine operation and maintenance method provided by the embodiments of the present application can be included in the application program 10022.

[0077] The method disclosed by the embodiments of the present application can be applied to the processor 1001 or implemented by the processor 1001. The processor 1001 can be an integrated circuit chip having a signal processing capability. In the implementation process, each step of the above method can be completed by an integrated logic circuit or an instruction in the form of software in the processor 1001. The processor 1001 can be a general processor, a digital signal processor (DSP), or other programmable logic device, discrete gate or transistor logic device, discrete hardware component, etc. The processor 1001 can implement or execute the disclosed methods, steps and logic block diagrams in the embodiments of the present application. The general processor 1001 can be a microprocessor or any conventional processor, etc. In combination with the steps of the accessory optimization method provided by the embodiments of the present application, the hardware decoding processor can be directly embodied to complete the execution, or the hardware and software modules in the decoding processor can be combined to complete the execution. The software module can be located in a storage medium, which is located in a memory. The processor reads the information in the memory and combines the hardware to complete the steps of the foregoing method.

[0078] In the exemplary embodiments, the terminal 1000 can be one or more application specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), or the like for executing the foregoing method.

[0079] Those of ordinary skill in the art can understand that all or part of the steps of the foregoing method embodiments can be completed by computer program related hardware. The foregoing computer program can be stored in a computer readable storage medium. When the program is executed, the steps of the foregoing method embodiments are executed; and the foregoing storage medium includes ROM, RAM, magnetic or optical disc, and various media capable of storing program codes.

[0080] In the embodiments provided in the present application, the computer readable and writable storage medium can include read-only memory, random access memory, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage device, flash memory, U disk, mobile hard disk, or any other medium capable of storing desired program code in the form of instructions or data structures and capable of being accessed by a computer. In addition, any connection can be appropriately referred to as a computer readable medium. For example, if instructions are sent from a website, server or other remote source using a coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL) or wireless technology such as infrared, radio and microwave, the coaxial cable, fiber optic cable, twisted pair, DSL or wireless technology such as infrared, radio and microwave is included in the definition of the medium. However, it should be understood that the computer readable and writable storage medium and the data storage medium do not include connections, carriers, signals or other transitory media, but are intended for non-transitory, tangible storage media. As used in the application, magnetic disks and optical disks include compact discs (CD), laser discs, optical discs, digital versatile discs (DVD), floppy disks and Blu-ray discs, in which magnetic disks typically magnetically copy data, and optical disks optically copy data with a laser.

[0081] In summary, the marine engine operation and maintenance method, system and terminal based on vibration signals and LLM of the present application first collect the original vibration signals and crank angle signals of each ship engine vibration measuring point, preprocess the original vibration signals and count typical features, and store the related signals in the recent preset time period. Then, based on the typical feature identification of the abnormal measuring point, a parameter data set is constructed for the abnormal measuring point. Then, the data set is input into the large language model in the vertical field, different tools are called, and the knowledge database is searched to output a fault analysis report. When the operation and maintenance personnel inquire about the analysis results, the model also answers through the knowledge database. The present application combines the cognitive reasoning of the large language model (LLM) with the tool calling ability of the Function Calling technology, realizes automatic diagnosis and natural language interaction, and is deployed on a specific optimized device. In the scenario of poor network and high communication cost of ocean-going ships, most problems can be solved without expert support, reducing operation and maintenance costs, and the use threshold is low. The operation and maintenance user interacts with the large language model to obtain various operation and maintenance information and details. Therefore, the present application effectively overcomes the various shortcomings in the prior art and has high industrial utilization value.

[0082] The above embodiments merely illustrate the principles of the present application and its effects, and are not intended to limit the present application. Any modification or change made by any person skilled in the art without departing from the spirit and scope of the present application shall be covered by the claims of the present application.

Claims

1. A marine engine operation and maintenance method based on vibration signals and LLM, characterized in that, The method comprises: Collecting original vibration signals and crank angle signals of each marine engine vibration measuring point; Pretreating the original vibration signals of each marine engine vibration measuring point and counting typical features, and simultaneously storing the original vibration signals and crank angle signals of each marine engine vibration measuring point in a preset time period; Identifying the abnormality of each marine engine vibration measuring point based on the counted typical features, and constructing a parameter data set for the marine engine vibration measuring point identified as abnormal; Inputting each parameter data set into a vertical domain large language model, calling different tools for analysis, and using a knowledge database for retrieval to output a fault analysis report, and when inputting natural language interactive information asked by an operation and maintenance personnel into the vertical domain large language model, using the knowledge database for retrieval and outputting corresponding answer information.

2. The vibration signal and LLM-based marine engine operation and maintenance method according to claim 1, characterized in that, The sampling frequency and sampling time of the original vibration signals and crank angle signals of each marine engine vibration measuring point are determined according to the engine speed range, cycle period and vibration characteristics of each marine engine vibration measuring point.

3. The vibration signal and LLM-based marine engine operation and maintenance method according to claim 2, characterized in that, Each marine engine vibration measuring point comprises an engine cylinder head, an engine cylinder sleeve and an engine frame, wherein the original vibration signals of each marine engine vibration measuring point are collected by vibration sensors arranged on the engine cylinder head, the engine cylinder sleeve and the engine frame respectively, and the crank angle signals are collected by a crank angle collecting device arranged at an engine flywheel disc.

4. The vibration signal and LLM-based marine engine operation and maintenance method according to claim 3, characterized in that, The pretreatment of the original vibration signals of each marine engine vibration measuring point and the counting of typical features comprise: Filtering and trend item removal are performed on the original vibration signals of each marine engine vibration measuring point; Time domain features and frequency domain features are extracted from the vibration signals after filtering and trend item removal, and typical features of each marine engine vibration measuring point are obtained and stored.

5. The vibration signal and LLM-based marine engine operation and maintenance method according to claim 4, characterized in that, The abnormality identification of each marine engine vibration measuring point based on the counted typical features and the construction of a parameter data set for the marine engine vibration measuring point identified as abnormal comprise: The typical features of each marine engine vibration measuring point are input into the constructed abnormality detection model for abnormality identification, and the abnormality identification results of each marine engine vibration measuring point are obtained; Based on the abnormality identification results, the abnormal marine engine vibration measuring points are locked, and the original vibration signals, typical features, crank angle signals and external system parameters of the corresponding marine engine vibration measuring points in a preset time period are integrated to obtain the parameter data set of each marine engine vibration measuring point.

6. The vibration signal and LLM-based marine engine operation and maintenance method according to claim 5, characterized in that, The input of each parameter data set into a vertical domain large language model, the calling of different tools for analysis and the use of a knowledge database for retrieval to output a fault analysis report comprise: In the tool set, the adaptive tools are sequentially called to analyze each parameter data set input into the vertical domain large language model to obtain preliminary analysis results, and then relevant knowledge is retrieved from the knowledge database according to the preliminary analysis results, and then the adaptive tools are again called from the tool set to analyze the retrieved knowledge, and based on the new analysis results, the retrieval in the knowledge database is continued, and through cyclic iteration, perfect fault analysis results are obtained; Based on the fault analysis results, a fault analysis report is generated and a visual chart is drawn for output.

7. The vibration signal and LLM-based marine engine operation and maintenance method according to claim 6, characterized in that, The tool set comprises analysis tools respectively configured based on each marine engine vibration measuring point and simulation tools respectively configured based on each fault scenario; The vertical domain large language model can call corresponding analysis tools for analysis according to the data of the measuring point, and when detecting that the data meets a fault scenario, call corresponding simulation tools for analysis.

8. The vibration signal and LLM-based marine engine operation and maintenance method according to claim 1, characterized in that, The vertical domain large language model is optimized by the information fed back by the operation and maintenance personnel according to the fault analysis report and the answer information.

9. A marine engine operation and maintenance system based on vibration signals and LLM, characterized in that, The system comprises: A data acquisition module is configured to acquire original vibration signals and crank angle signals of each marine engine vibration measuring point; A feature statistics module is connected to the data acquisition module and configured to preprocess the original vibration signals of each marine engine vibration measuring point and count typical features; A storage module is connected to the feature statistics module and the data acquisition module and comprises a time series database and an in-memory database, which are respectively configured to long-term store the counted typical features and short-term store the original vibration signals and crank angle signals of each marine engine vibration measuring point in a preset time period; An abnormality identification module is connected to the storage module and configured to identify abnormalities of each marine engine vibration measuring point based on the counted typical features, and construct a parameter data set for the marine engine vibration measuring point identified as abnormal; A vertical domain large language model analysis and interaction module is connected to the abnormality identification module and configured to input each parameter data set into a vertical domain large language model, call different tools for analysis, and use a knowledge database to search and output a fault analysis report, and when natural language interaction information is input into the vertical domain large language model by the operation and maintenance personnel according to the fault analysis result, use the knowledge database to search and output corresponding answer information.

10. An electronic terminal, characterized in that It comprises: One or more memories and one or more processors; The one or more memories are configured to store a computer program; The one or more processors are connected to the memories and configured to run the computer program to perform the method in claim 8.

Images