A method, system, storage medium, and equipment for detecting abnormalities in marine diesel engines.

Abstract

This invention provides a method, system, storage medium, and device for detecting anomalies in marine diesel engines. The method includes collecting multi-dimensional operating data of the diesel engine; preprocessing the data to obtain standardized time-segment data; extracting candidate anomaly events; constructing a similarity measurement mechanism between time series and obtaining a similarity matrix; using an unsupervised clustering method to aggregate behavioral patterns of the candidate anomaly events and identify anomaly event clusters; structuring the anomaly event clusters into semantic anomaly events with clear meanings and field definitions; organizing the semantic anomaly events into structured event objects and outputting them in a standard format. This invention can automatically extract potential anomaly events, possessing stronger morphological recognition capabilities, adaptability, and versatility. Its output results not only facilitate rapid source tracing and analysis by experts but also have good interface compatibility, significantly improving the intelligence level and engineering practical value of marine diesel engine condition monitoring systems.

Description

Technical Field

[0001] This invention relates to the field of marine diesel engine condition monitoring technology, and in particular to a method, system, storage medium and equipment for detecting abnormalities in marine diesel engines. Background Technology

[0002] Marine diesel engines are core equipment operating under high loads for extended periods, and their operational status directly impacts navigation safety and energy efficiency. Current mainstream monitoring methods largely rely on the following approaches: 1) Threshold-based alarm mechanisms, which set upper and lower limits for sensor parameters to determine if boundaries have been exceeded in real time, but cannot identify more complex anomalies such as response lag or abnormal trends; 2) Rule-based systems based on expert experience, which rely on manually set rules for judgment, have poor generalization capabilities, and are difficult to adapt to complex operating conditions and multiple engine models; 3) Offline manual backtesting analysis, suitable for post-event diagnosis, but suffers from poor real-time performance and low efficiency; 4) Data-driven methods, such as clustering and classification models, are used for modeling, but are mainly limited to anomaly detection and largely rely on labeled data, making it difficult to automatically extract abnormal behavior from unlabeled data. With the increasing demand for intelligent and remote ship maintenance, diesel engine anomaly detection systems are playing an increasingly prominent role in ensuring safe ship operation. However, most existing methods rely on fixed threshold judgments or expert rule systems. While they can achieve anomaly identification under some typical operating conditions, their ability to model the deep structural relationships of multidimensional sensor data is limited. Especially when facing complex operating states and unlabeled data, their accuracy and adaptability in anomaly identification are significantly insufficient. More importantly, current system outputs are generally presented in the form of "is it abnormal?" or simple alarms, lacking a complete expression with structured semantics such as time range, type classification, and behavioral characteristics. This makes it difficult for experts to effectively understand and track the detection results, and also makes it difficult to serve as an input interface for intelligent analysis systems, limiting the interpretability and intelligent expansion capabilities of the system. Summary of the Invention

[0003] In view of the above-mentioned problems in the prior art, the present invention provides a method, system, storage medium and device for detecting abnormalities in marine diesel engines, in order to solve the technical problems that the accuracy and adaptability of abnormal identification in the prior art are greatly insufficient, and the current system output results are generally presented in the form of "whether it is abnormal" or simple alarm, lacking a complete expression of structured semantics.

[0004] This invention provides a method for detecting abnormalities in marine diesel engines, comprising the following steps:

[0005] Step S1: Collect multi-dimensional operating data of the diesel engine, including speed, temperature, pressure and torque;

[0006] Step S2: Preprocess the data by using a sliding window to segment the time series and obtain standardized time segment data;

[0007] Step S3: Extract standardized time segment data with dynamic change characteristics as candidate anomalous events;

[0008] Step S4: Construct a similarity measurement mechanism between time series and obtain the similarity matrix;

[0009] Step S5: Based on the similarity matrix, an unsupervised clustering method is used to aggregate behavioral patterns of candidate abnormal events and identify abnormal event clusters.

[0010] Step S6: For the clusters of abnormal events identified after clustering, combine factors including the waveform features of the cluster center, the main change dimensions, the duration of the abnormality, and the similarity score, and structure them into semantic abnormal events with clear meaning and field definitions. The semantic abnormal events include the abnormality type, the dominant variable, the start and end time of the event, and the score value.

[0011] Step S7: Organize semantic exception events into structured event objects and output them in a standard format.

[0012] In one embodiment, the preprocessing in step S2 includes format normalization, noise reduction, missing data completion, unit conversion, and normalization.

[0013] In one embodiment, the similarity measurement mechanism employs a method including a dynamic time warping algorithm or a distance function algorithm with temporal alignment capabilities.

[0014] In one embodiment, the unsupervised clustering method is the PIC algorithm, DBSCAN algorithm, k-means algorithm, or a clustering algorithm based on density and shape analysis.

[0015] In one embodiment, the similarity matrix acquisition and behavior pattern aggregation and recognition process are performed through a feature space constructed by a deep neural network.

[0016] In one embodiment, the anomaly type is used to describe the semantic behavior pattern of anomaly events, and its classification is generated based on one or a combination of two strategies: rule template matching and data-driven classifier.

[0017] In addition, embodiments of the present invention also provide a marine diesel engine anomaly detection system, comprising:

[0018] A data acquisition module is used to acquire multi-source signal data of the marine diesel engine under various operating conditions in real time to obtain multi-dimensional time series data;

[0019] A preprocessing module is used to receive raw time series data from the data acquisition module, perform format normalization, noise reduction, missing data completion, unit conversion and normalization on the data, and divide the continuous time series into fixed-length segments through a sliding window mechanism to form standardized time segment data.

[0020] An event extraction module is used to extract candidate abnormal events;

[0021] A similarity modeling module is provided, which is used to evaluate the temporal morphological similarity between candidate abnormal events and construct a similarity matrix for cluster analysis.

[0022] A clustering analysis module, which uses an unsupervised clustering algorithm to automatically group events and identify abnormal event clusters that are structurally significantly different from the normal operating mode;

[0023] A semantic construction module is used to structure the clusters of abnormal events identified after clustering into semantic abnormal events with clear meaning and field definitions;

[0024] A result output module is provided, which is used to output structured event objects in a standard format.

[0025] In addition, embodiments of the present invention also provide a computer-readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the marine diesel engine anomaly detection method described in any embodiment of the present invention.

[0026] In addition, embodiments of the present invention also provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the marine diesel engine anomaly detection method described in any embodiment of the present invention.

[0027] Compared with existing technologies, the beneficial effects of the marine diesel engine anomaly detection method, system, storage medium, and equipment provided by the embodiments of the present invention are as follows: The embodiments of the present invention can effectively overcome the limitations of traditional diesel engine monitoring methods in terms of operating condition adaptability, anomaly granularity identification, and result interpretability. In the face of complex ship operating conditions and unlabeled large-scale multidimensional data scenarios, it can automatically extract potential abnormal events and organize them into structured semantic results with time intervals, behavior types, key feature indicators, and similarity scores. Compared with traditional methods that rely on fixed thresholds and expert experience rules, the embodiments of the present invention have stronger morphological recognition capabilities, adaptability, and versatility. Its output results not only facilitate rapid source tracing and analysis by experts, but also have good interface compatibility, and can be directly connected to intelligent systems such as large language models and expert diagnostic engines to achieve closed-loop fusion between anomaly detection, semantic understanding, and intelligent decision-making, thereby significantly improving the intelligence level and engineering practical value of the marine diesel engine condition monitoring system. Attached Figure Description

[0028] Figure 1 A flowchart illustrating a method for detecting abnormalities in a marine diesel engine, provided by an embodiment of the present invention;

[0029] Figure 2 This is a schematic diagram of the composition of a marine diesel engine anomaly detection system provided in an embodiment of the present invention. Detailed Implementation

[0030] To enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0031] Various embodiments and features of this application are described herein with reference to the accompanying drawings.

[0032] These and other features of this application will become apparent from the following description of preferred forms of embodiments given as non-limiting examples, with reference to the accompanying drawings.

[0033] It should also be understood that although this application has been described with reference to some specific examples, those skilled in the art can certainly implement many other equivalent forms of this application, which have the features described in the claims and are therefore all within the scope of protection defined herein.

[0034] The above and other aspects, features and advantages of this application will become more apparent when taken in conjunction with the accompanying drawings and in view of the following detailed description.

[0035] Specific embodiments of this application are described below with reference to the accompanying drawings; however, it should be understood that the claimed embodiments are merely examples of this application, which can be implemented in various ways. Well-known and / or repeated functions and structures are not described in detail to ascertain the true intent based on the user's historical operations, and to avoid unnecessary or redundant details that would obscure this application. Therefore, the specific structural and functional details claimed herein are not intended to be limiting, but merely serve as the basis and representative basis for the claims to teach those skilled in the art to use this application in various ways with substantially any suitable detailed structure.

[0036] This specification may use the phrases “in one embodiment,” “in another embodiment,” “in yet another embodiment,” or “in other embodiments,” all of which may refer to one or more of the same or different embodiments according to this application.

[0037] The principles and features of the present invention are described below with reference to the accompanying drawings. The embodiments described are for illustrative purposes only and are not intended to limit the scope of the invention. The following description, in conjunction with... Figure 1-2 The preferred embodiments of the present invention will be described in further detail below:

[0038] like Figure 1 As shown, this embodiment of the invention provides a method for detecting abnormalities in a marine diesel engine, comprising the following steps:

[0039] Step S1: Collect multi-dimensional operating data of the diesel engine, including speed, temperature, pressure and torque;

[0040] Step S2: Preprocess the data by using a sliding window to segment the time series and obtain standardized time segment data;

[0041] Step S3: Extract standardized time segment data with dynamic change characteristics as candidate anomalous events;

[0042] Step S4: Construct a similarity measurement mechanism between time series and obtain the similarity matrix;

[0043] Step S5: Based on the similarity matrix, an unsupervised clustering method is used to aggregate behavioral patterns of candidate abnormal events and identify abnormal event clusters.

[0044] Step S6: For the clusters of abnormal events identified after clustering, combine factors including the waveform features of the cluster center, the main change dimensions, the duration of the abnormality, and the similarity score, and structure them into semantic abnormal events with clear meaning and field definitions. The semantic abnormal events include the abnormality type, the dominant variable, the start and end time of the event, and the score value.

[0045] Step S7: Organize semantic exception events into structured event objects and output them in a standard format.

[0046] Specifically, this invention proposes a method for detecting anomalies in marine diesel engines based on time-series semantic analysis. The aim is to identify potential abnormal behaviors from large-scale unlabeled operational data and output the detection results in a structured semantic form, thereby improving the operational safety and intelligent maintenance level of marine power systems. This invention fully considers the complex characteristics of nonlinearity, multiple operating conditions, and multiple variables in the operation of marine diesel engines, and designs a highly adaptable and interpretable anomaly detection technology system.

[0047] This invention first collects multi-dimensional operating data of a diesel engine, including but not limited to key parameters such as speed, temperature, pressure, and torque. Time series data is segmented using a sliding window, and the raw data is standardized and denoised to extract data segments with dynamic characteristics as candidate anomalies. Subsequently, a similarity measurement mechanism between time series is constructed to measure the morphological similarity of candidate events. The preferred similarity measurement method is Dynamic Time Warping (DTW), but Euclidean distance, edit distance, or other distance functions with temporal alignment capabilities can also be used.

[0048] After obtaining the similarity matrix, an unsupervised clustering method is further employed to aggregate behavioral patterns in the event fragments, thereby classifying and identifying abnormal patterns. The preferred clustering method is the Power Iteration Clustering (PIC) algorithm, but DBSCAN, k-means, or other clustering techniques based on density and shape analysis can also be used. In practical implementation, this step can dynamically adjust the clustering parameters according to different models or operating conditions of the marine diesel engine to improve the robustness of anomaly detection.

[0049] For the abnormal clusters identified after clustering, this invention introduces a semantic tag generation step. Combining factors such as cluster center waveform features, main change dimensions, anomaly duration, and similarity scores, it constructs anomaly event description units with structured semantics. Each anomaly event includes at least the event start and end time, anomaly type, key parameter indicators, and a scoring field. This structured semantic result not only facilitates intuitive analysis by experts and system backtracking, but also serves as input for integration with large language models, expert systems, or remote fault analysis platforms, enabling intelligent interpretation and utilization of anomaly detection results.

[0050] Furthermore, in extended embodiments, the similarity modeling and clustering process can also be performed using a feature space constructed through a deep neural network, such as using an autoencoder for nonlinear feature compression, combined with a clustering algorithm to complete unsupervised anomaly pattern recognition. The technical framework provided by this invention has good versatility and scalability, and is suitable for diesel engine condition monitoring needs under various ship types and complex navigation conditions.

[0051] In addition, this invention also provides a marine diesel engine anomaly detection system, which can be deployed on a marine diesel engine monitoring system or an edge intelligent processing platform. The system consists of multiple functional modules, including data acquisition, preprocessing, anomaly event extraction, similarity calculation, cluster analysis, semantic construction, and result output.

[0052] (1) The main functions of the data acquisition module are:

[0053] Multi-source signal data of marine diesel engines under various operating conditions are collected in real time, including various types of signals such as speed, pressure, temperature, and vibration, to obtain multi-dimensional time series data. Table 1 lists typical input variables.

[0054] Table 1 Key Operating Parameters of Marine Diesel Engines

[0055] 1 Engine_Speed rpm Spindle speed 2 Fuel_Pressure MPa Fuel system pressure 3 Oil Temperature ℃ Lubricating oil temperature 4 Exhaust_Temperature ℃ Exhaust temperature of each cylinder 5 Vibration_XYZ <![CDATA[m / s 2 ]]> Engine base vibration (3-axis)

[0056] (2) The main functions of the preprocessing module are:

[0057] The system receives raw time-series data from the acquisition module and performs formatting, noise reduction, missing data completion, unit conversion, and normalization. Then, a sliding window mechanism is used to segment the continuous time series into fixed-length segments, forming the basic data units for subsequent analysis. With a window length of w seconds and a step size of δ seconds, the total data D is divided into several subsequences {d1, d2, ..., d...}. n Each subsequence contains multiple synchronous sampling channels. This module outputs a standardized set of short-time, multivariate time-series sequences, providing a foundation for event recognition and behavior modeling.

[0058] (3) The main functions of the event extraction module are:

[0059] Candidate segments with anomalous characteristics are identified from the pre-processed time series; these segments are data that may differ from routine operating conditions in terms of behavior. Event extraction employs a rule-based approach based on multivariate statistical features, combined with engineering heuristics for screening. For each time segment d... i Calculate the following characteristics of its key variables:

[0060] maximum slope Determine if a mutation exists;

[0061] Number of extreme values ​​N peak (x): Determine whether a drastic fluctuation has occurred;

[0062] Standard deviation σ(x): indicates the stability of the variable;

[0063] Main variable amplitude ratio:

[0064] If any of the above features exceeds a set threshold (as given by an empirical model), then the segment d is... i Record these as candidate exception events. The output of this module is a set of candidate events E = {e1, e2, ..., e...} k This data is used for subsequent similarity calculations and cluster analysis. This method can filter out a large number of uninformative and stable running segments, improving the efficiency and quality of subsequent clustering.

[0065] (4) The main functions of the similarity modeling module are:

[0066] This module is responsible for evaluating the temporal morphological similarity between candidate events and constructing a metric space for cluster analysis. Dynamic Time Warping (DTW) is used as the core metric method for any two events e. i e j Pairing and comparing multivariate sequences, calculate their minimum deformable alignment path cost:

[0067]

[0068] Where π represents the valid alignment path, and x(m) represents the m-th time point in the event sequence. This distance reflects the morphological similarity of events under time-axis deformation tolerance. A similarity matrix is ​​constructed by calculating the similarity of all events pairwise. Its elements are:

[0069]

[0070] Where σ is the width parameter set empirically.

[0071] (5) The main functions of the cluster analysis module are:

[0072] The clustering module receives the similarity matrix output by the previous module.

[0073] In a preferred embodiment, the Power Iteration Clustering (PIC) method is employed. This method first constructs a normalized similarity matrix A = S, and then performs power iteration calculations:

[0074] v t+1 =A·v t

[0075] Where v0 is a random non-zero initial vector.

[0076] Once the iteration converges, the principal components of v are used as clustering features, combined with k-means or spectral clustering methods to complete the final classification. The number of clusters k can be automatically estimated using the elbow rule or density analysis.

[0077] The final output includes: the cluster label of each event; statistics such as density and center curve of each cluster; and low-density and morphologically aberrant clusters are marked as anomalous clusters.

[0078] The core objective of this module is to proactively discover morphological outliers from unlabeled historical data in an unsupervised manner, significantly reducing reliance on human experience and labeled data.

[0079] (6) The main functions of the semantic construction module are:

[0080] The anomaly semantic construction module is one of the core components of this invention. Its function is not only to mark anomalies, but also to extract, interpret, and organize understandable semantic information of anomalous events from time series data, achieving a leap from numerical anomalies to semantic anomalies. This module further structures the anomalous event fragments identified in the previous stage of cluster analysis into semantic anomalous events with clear meanings and defined fields, including anomaly type, dominant variable, event start and end times, and score values, providing a unified input format for expert analysis and intelligent system calls.

[0081] The processing flow of this module includes three core steps: abnormal behavior classification (event_type), dominant variable identification (main_variable), and abnormal score calculation and structured output. The generation methods for each field are as follows:

[0082] 1) Classification method of exception type (event_type)

[0083] Anomaly types describe the semantic behavior patterns of events, such as "acceleration hysteresis," "power fluctuations," and "cooling response anomalies." Their generation is based on one or a combination of the following two strategies:

[0084] Rule template matching: A pre-defined library of anomaly types is used, with each type defining a set of typical characteristics (such as trends, response delays, fluctuation patterns, etc.). When the main variable of an anomaly event matches a certain type template with a high degree of similarity, that label can be assigned.

[0085] For example:

[0086] If the variable response is significantly delayed and the slope of change is low, it matches "hysteresis response";

[0087] If a variable oscillates frequently and has a large standard deviation, it matches the "abnormal fluctuation" rule.

[0088] Data-driven classifier (optional): Supervised learning training is performed on historically classified data to form a multi-classification model based on feature vectors. During deployment, the type label is predicted directly based on the input features. Suitable for the stage after the accumulation of big data.

[0089] 2) Methods for identifying the dominant variable (main_variable)

[0090] The dominant variable is used to identify the sensor variable that has a dominant influence on the behavior of anomalies. To quantify the influence of each variable in anomalies, this invention introduces the concept of anomaly impact factor, which comprehensively considers the two dimensions of variable fluctuation amplitude and rate of change.

[0091] For each variable x i Its impact factor is defined as follows:

[0092] Impact(x i )=ω1·Range(x i )+ω2·Grad(x i )

[0093] in, Indicates the fluctuation range of the variable. The average rate of change of a variable (i.e., the average absolute slope), x i Let T be the time series of the i-th variable during the time period of the abnormal event. e ω1 and ω2 are the duration of the current abnormal event (number of sampling points), and are weighting coefficients that satisfy ω1 + ω2 = 1, used to adjust the relative contribution of amplitude and rate.

[0094] The system for each variable x i Calculate its Impact(x) i The values ​​are calculated, and the largest one is selected as the dominant variable for the current abnormal event.

[0095]

[0096] If the scores of multiple variables are close, the system can record the ranking of candidate main variables, or output two dominant variables side by side to support subsequent manual analysis or expert decision-making.

[0097] 3) How to calculate the anomaly score

[0098] The system quantitatively scores the significance of anomalies by comprehensively considering the duration, amplitude fluctuations, and trends of the events. The scoring function is as follows:

[0099] Score = α·T norm +β·S amp +γ·V grad

[0100] Among them, T norm =T e / T max T is the normalized value of the event duration. e T represents the duration of the event. max The maximum allowable reference duration;

[0101] S amp = (max(x) - min(x)) / A ref The amplitude ratio represents the ratio of the maximum fluctuation amplitude of the main variable within the event segment to the global reference value A. ref The ratio;

[0102] The average rate of change (slope) of the main variable in an event is used to quantify the degree of abrupt change in trend;

[0103] α, β, and γ are weighting coefficients used to adjust the contribution of each item to the score. Generally, they satisfy α + β + γ = 1. Their values ​​can be set through training with historical samples or preset based on actual engineering experience.

[0104] The calculated score ranges from [0,1] and can be used as a reference for the credibility of an event or the intensity of an anomaly.

[0105] 4) Generation and output of structured semantic results

[0106] The above fields are combined into a structured event object for easy unified storage, remote transmission, and multi-system interface calls. The output format is as follows:

[0107] Table 2 Definition of Semantic Fields for Abnormal Events

[0108] event_id String Unique ID of Abnormal Event event_type String Anomaly categories (such as hysteresis, fluctuation, etc.) main_variable String The dominant variable corresponding to the anomaly start_time Date and Time abnormal event start time end_time Date and Time End time of abnormal event score number Abnormal scoring results, range [0,1]

[0109] The output sample is as follows:

[0110]

[0111] (7) The main functions of the result output module are:

[0112] This module outputs structured exception results in a standard format (such as JSON or database records) for remote platform access. Supported integration:

[0113] Expert systems are used to support decision-making.

[0114] Large language models (such as GPT) → automatic report generation;

[0115] Operations and Maintenance Platform → Fault Prediction and Maintenance Scheduling.

[0116] The module also supports setting confidence thresholds to perform event filtering and feedback processing, forming a closed-loop analysis process.

[0117] In addition, embodiments of the present invention also provide a computer-readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the marine diesel engine anomaly detection method described in any embodiment of the present invention.

[0118] In addition, embodiments of the present invention also provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the marine diesel engine anomaly detection method described in any embodiment of the present invention.

[0119] The above embodiments are merely exemplary embodiments of the present invention and are not intended to limit the present invention. The scope of protection of the present invention is defined by the claims. Those skilled in the art can make various modifications or equivalent substitutions to the present invention within its spirit and scope of protection, and such modifications or equivalent substitutions should also be considered to fall within the scope of protection of the present invention.

Claims

1. A method for detecting abnormalities in marine diesel engines, characterized in that, Includes the following steps: Step S1: Collect multi-dimensional operating data of the diesel engine, including speed, temperature, pressure and torque; Step S2: Preprocess the data by using a sliding window to segment the time series and obtain standardized time segment data; Step S3: Extract standardized time-segment data with dynamic change characteristics as candidate anomalous events, as detailed below. Candidate segments with anomalous characteristics are identified from preprocessed time series. Event extraction employs a rule-based method based on multivariate statistical features, combined with engineering heuristics for filtering. For each time segment... Calculate the following characteristics of its key variables: maximum slope : Determine if a mutation exists; Number of extreme values : Determine whether drastic fluctuations have occurred; Standard deviation : Indicates variable stability; Main variable amplitude ratio: If any of the above features exceeds a set threshold, then the segment will be... Recorded as a candidate exception event; Step S4: Construct a similarity measurement mechanism between time series and obtain the similarity matrix; Step S5: Based on the similarity matrix, an unsupervised clustering method is used to aggregate behavioral patterns of candidate abnormal events and identify abnormal event clusters. Step S6: For the clusters of anomalous events identified after clustering, and combining factors including cluster center waveform features, main change dimensions, anomaly duration, and similarity score, they are structured into semantic anomalous events with clear meaning and field definitions. These semantic anomalous events include anomaly type, dominant variable, event start and end times, and score value. The dominant variable is used to identify the sensor variable that has a dominant influence on the behavior of anomalies. By introducing an anomaly impact factor, the variable's fluctuation amplitude and rate of change are considered comprehensively. For each variable Its impact factor is defined as follows: in, Indicates the fluctuation range of the variable. This represents the average rate of change of the variable. For the first The time series of each variable during the period of the abnormal event. This represents the duration of the current abnormal event (number of sampling points). Let be the weighting coefficient, satisfying This is used to adjust the relative contribution of amplitude and rate. The system for each variable Calculate its The values ​​are determined, and the largest one is selected as the dominant variable for the current anomaly event. If the scores of multiple variables are close, the ranking of candidate main variables can be recorded, or two dominant variables can be output side by side to support subsequent manual analysis or expert decision-making. Step S7: Organize semantic exception events into structured event objects and output them in a standard format.

2. The method for detecting abnormalities in a marine diesel engine according to claim 1, characterized in that: The preprocessing described in step S2 includes format normalization, noise reduction, missing data completion, unit conversion, and normalization.

3. The method for detecting abnormalities in a marine diesel engine according to claim 1, characterized in that: The similarity measurement mechanism employs methods including dynamic time warping algorithms or distance function algorithms with temporal alignment capabilities.

4. The method for detecting abnormalities in a marine diesel engine according to claim 1, characterized in that: The unsupervised clustering method is the PIC algorithm, DBSCAN algorithm, k-means algorithm, or a clustering algorithm based on density and shape analysis.

5. The method for detecting abnormalities in a marine diesel engine according to claim 1, characterized in that: The similarity matrix is ​​obtained and the behavior pattern aggregation and recognition process is carried out through a feature space constructed by a deep neural network.

6. The method for detecting abnormalities in a marine diesel engine according to claim 1, characterized in that: The anomaly type is used to describe the semantic behavior pattern of the abnormal event. Its classification is generated based on one or a combination of two strategies: rule template matching and data-driven classifier.

7. A marine diesel engine anomaly detection system, characterized in that, include: A data acquisition module is used to acquire multi-source signal data of the marine diesel engine under various operating conditions in real time to obtain multi-dimensional time series data; A preprocessing module is used to receive raw time series data from the data acquisition module, perform format normalization, noise reduction, missing data completion, unit conversion and normalization on the data, and divide the continuous time series into fixed-length segments through a sliding window mechanism to form standardized time segment data. An event extraction module is used to extract candidate abnormal events. The extraction process is as follows: Candidate segments with anomalous characteristics are identified from preprocessed time series. Event extraction employs a rule-based method based on multivariate statistical features, combined with engineering heuristics for filtering. For each time segment... Calculate the following characteristics of its key variables: maximum slope : Determine if a mutation exists; Number of extreme values : Determine whether drastic fluctuations have occurred; Standard deviation : Indicates variable stability; Main variable amplitude ratio: If any of the above features exceeds a set threshold, then the segment will be... Recorded as a candidate exception event; A similarity modeling module is provided, which is used to evaluate the temporal morphological similarity between candidate abnormal events and construct a similarity matrix for cluster analysis. A clustering analysis module, which uses an unsupervised clustering algorithm to automatically group events and identify abnormal event clusters that are structurally significantly different from the normal operating mode; A semantic construction module is used to structure the clusters of anomalous events identified after clustering into semantic anomalous events with clear meanings and field definitions. These semantic anomalous events include anomaly type, dominant variable, event start and end times, and score values. The dominant variable is used to identify the sensor variable that has a dominant influence on the behavior of anomalies. By introducing an anomaly impact factor, the variable's fluctuation amplitude and rate of change are considered comprehensively. For each variable Its impact factor is defined as follows: in, Indicates the fluctuation range of the variable. This represents the average rate of change of the variable. For the first The time series of each variable during the period of the abnormal event. This represents the duration of the current abnormal event (number of sampling points). Let be the weighting coefficient, satisfying This is used to adjust the relative contribution of amplitude and rate. The system for each variable Calculate its The values ​​are determined, and the largest one is selected as the dominant variable for the current anomaly event. If the scores of multiple variables are close, the ranking of candidate main variables can be recorded, or two dominant variables can be output side by side to support subsequent manual analysis or expert decision-making. A result output module is provided, which is used to output structured event objects in a standard format.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by the processor, it implements claim 1. The method for detecting abnormalities in marine diesel engines as described in any one of the six claims.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements claim 1. The method for detecting abnormalities in marine diesel engines as described in any one of the six claims.

Images