Coupling type ship power system fault diagnosis method and device and electronic equipment

Abstract

The invention provides a coupled ship power system fault diagnosis method and device and electronic equipment, and relates to the field of data processing. According to the method, multi-source data of a coupled ship power system is acquired through a heterogeneous sensing array, and enhanced sensing flow is generated by using data gap completion and scene enhancement; performing redundancy removal, robustness and semantic segmentation on the enhanced perception stream, and extracting time sequence fragments of equipment fluctuation, energy rheology and operation actions to form an event feature sequence; carrying out recombination comparison by combining the hidden layer evolution script and weak mode recognition, and outputting a hidden coupling candidate link set; mapping the link set to a three-dimensional hull space and a control interface, and generating a spatialized coupling light band scene; and finally, generating a root cause analysis report by means of the fault analysis tree according to the key optical band nodes. By implementing the technical scheme provided by the invention, the fault diagnosis accuracy of the coupled ship power system can be improved.

Description

Technical Field

[0001] This application relates to the technical field of data processing, specifically to a method, apparatus, and electronic equipment for diagnosing faults in a coupled marine power system. Background Technology

[0002] With the development of technology, coupled marine propulsion systems are rapidly evolving towards higher efficiency, intelligence, and greener operation. However, as the architecture of propulsion systems becomes more complex, the interactions between equipment are becoming increasingly close, and the operating status is exhibiting a high degree of dynamism and uncertainty.

[0003] Currently, existing fault diagnosis methods generally adopt an equipment-centric paradigm. Their core characteristic is focusing on a single device or local subsystem, identifying potential faults by monitoring discrete anomalies in its operating parameters. While these methods are effective in detecting single-point faults, they lack the ability to comprehensively integrate multi-source heterogeneous data across the entire ship, failing to fully analyze the coupling relationships between devices. This limits the diagnostic scope to the single-device level, resulting in low accuracy in diagnosing faults in coupled marine propulsion systems.

[0004] Therefore, there is an urgent need for a coupled method, device, and electronic equipment for diagnosing faults in marine propulsion systems. Summary of the Invention

[0005] This application provides a method, apparatus, and electronic device for fault diagnosis of coupled marine propulsion systems, which facilitates improving the accuracy of fault diagnosis of coupled marine propulsion systems.

[0006] The first aspect of this application provides a fault diagnosis method for a coupled marine propulsion system. The method includes: acquiring raw multi-source data stream packets sent by a heterogeneous sensor array to the coupled marine propulsion system; generating an enhanced perception stream based on the raw multi-source data stream packets using data gap completion technology and scene enhancement technology; performing redundancy removal, robustness enhancement, and semantic segmentation on the enhanced perception stream, and extracting time-series segments of equipment fluctuations, energy rheology, and operational actions to generate an event-based feature sequence; recombining and comparing the event-based feature sequence using a latent layer evolution script and a weak pattern recognition algorithm to output a latent coupling candidate link set; mapping the latent coupling candidate link set to a three-dimensional hull space and control interface to generate a spatialized coupled light strip scene; and generating a root cause analysis report of the coupled marine propulsion system based on key light strip nodes in the spatialized coupled light strip scene using a fault analysis tree.

[0007] A second aspect of this application provides a fault diagnosis device for a coupled marine propulsion system. The device includes an acquisition module and a processing module. The acquisition module acquires raw multi-source data stream packets sent by a heterogeneous sensor array to the coupled marine propulsion system. The processing module generates an enhanced sensing stream based on the raw multi-source data stream packets using data gap completion and scene enhancement techniques. The processing module further performs redundancy removal, robustness enhancement, and semantic segmentation on the enhanced sensing stream, and extracts temporal segments of equipment fluctuations, energy rheology, and operational actions to generate an event-based feature sequence. The processing module also performs recombination and comparison of the event-based feature sequence using a latent layer evolution script and a weak pattern recognition algorithm to output a latent coupling candidate link set. The processing module further maps the latent coupling candidate link set to a three-dimensional hull space and control interface to generate a spatialized coupled light strip scene. The processing module also generates a root cause analysis report of the coupled marine propulsion system based on key light strip nodes in the spatialized coupled light strip scene using a fault analysis tree.

[0008] A third aspect of this application provides an electronic device including a processor, a memory, a user interface, and a network interface. The memory is used to store instructions, and both the user interface and the network interface are used to communicate with other devices. The processor is used to execute the instructions stored in the memory to cause the electronic device to perform the method described above.

[0009] A fourth aspect of this application provides a computer-readable storage medium storing instructions that, when executed, perform the method described above.

[0010] In summary, one or more technical solutions provided in this application have at least the following technical effects or advantages: Multi-source data is collected via heterogeneous sensor arrays, combined with data gap completion and scene enhancement to ensure data integrity and richness. Then, event-based feature sequences are generated through redundancy removal, robustness enhancement, and semantic segmentation to guarantee the accuracy and stability of feature extraction. Based on this, a set of latent coupling candidate links is output using latent layer evolutionary scripts and weak pattern recognition to reveal potential coupling relationships between devices. Furthermore, the link set is mapped to a spatialized coupled light strip scene, and causal reasoning is performed on key light strip nodes through fault analysis trees to generate root cause analysis reports, thereby achieving system-level health monitoring and risk warning for diesel-electric hybrid power systems. Therefore, this facilitates improved fault diagnosis accuracy for coupled marine power systems. Attached Figure Description

[0011] Figure 1 A flowchart illustrating a coupled marine propulsion system fault diagnosis method provided in an embodiment of this application; Figure 2 A schematic diagram of a coupled marine power system fault diagnosis device provided in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0012] Explanation of reference numerals in the attached figures: 21. Acquisition module; 22. Processing module; 31. Processor; 32. Communication bus; 33. User interface; 34. Network interface; 35. Memory. Detailed Implementation

[0013] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0014] In the description of the embodiments of this application, the words "for example" or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design that is described as "for example" or "for instance" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design options. Rather, the use of the words "for example" or "for instance" is intended to present the relevant concepts in a specific manner.

[0015] In the description of the embodiments of this application, the term "multiple" means two or more. For example, multiple systems means two or more systems, and multiple screen terminals means two or more screen terminals. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the indicated technical features. Thus, a feature defined with "first" or "second" may explicitly or implicitly include one or more of that feature. The terms "comprising," "including," "having," and variations thereof all mean "including but not limited to," unless otherwise specifically emphasized.

[0016] To address the aforementioned technical problems, this application provides a fault diagnosis method for coupled marine propulsion systems, referring to... Figure 1 , Figure 1 This is a flowchart illustrating a coupled marine propulsion system fault diagnosis method provided in an embodiment of this application. The method is applied to a server and includes steps S110 to S160, as follows: S110. Acquire the raw multi-source data stream packets sent by the heterogeneous sensor array for the coupled marine propulsion system.

[0017] Specifically, a server refers to a computing resource node deployed in a shore-based data center or a ship-shore integrated server room, responsible for data access, storage, scheduling, and quality inspection. It can be a physical host or a virtualized cluster, interconnected with edge computing units via Ethernet, 5G, or satellite leased lines. For example, a dual-socket x86 server acts as a data access node in a Kafka cluster, continuously handling data writing and verification from a coupled ship propulsion system. Acquisition refers to the server's continuous reception and storage of data through a subscription or pull mechanism, including long-connection subscriptions, message queue consumption, and batch retrieval. For example, a server subscribes to the topic "ship / propulsion / sensors / #", writes data to object storage and a time-series database in the order of arrival, and generates a receipt acknowledgment to achieve end-to-end traceability.

[0018] Heterogeneous sensor arrays refer to a collection of sensors with multiple physical quantities, sampling frequencies, and communication protocols that are collaboratively deployed around objects such as diesel engines, electric motors, gearboxes, generators, batteries, and propulsion shafts. They share common spatiotemporal calibration and consistent device identification. For example, triaxial accelerometers and temperature probes are deployed on the gearbox housing; individual voltage sampling boards and temperature and humidity sensors are deployed in the battery compartment; current and voltage transformers and shaft vibration probes are deployed on the generator side; and noise, wind speed, and wind direction sensors are deployed at environmental locations in the engine room. Coupled marine propulsion systems refer to an energy and torque coupled topology consisting of diesel engines, electric motors, generators, power conversion devices, gearboxes, power distribution buses, batteries, propulsion shafts, and propellers, embodying the dual interaction of mechanical and electrical coupling. For example, in a parallel system, the diesel engine and electric motor are coaxial and drive the propulsion shaft through the gearbox; in a series system, the diesel engine supplies power to the propulsion motor through the generator and converter and coordinates with battery energy management.

[0019] Sending refers to the process by which edge computing units or field gateways aggregate, serialize, and encapsulate sensor measurements before pushing them to the server, including retries, buffering, and quality of service control. For example, the edge computing unit aggregates multi-channel measurements in 100-millisecond time windows, serializes them using Protobuf, publishes them to the access gateway via MQTT or OPCUAsub, and then the gateway forwards them to the server-side message queue. Raw multi-source data stream packets refer to time-series data slices that have not undergone past redundancy, robustness, and fusion processing, but already possess unified timestamps and device identification specifications. They carry multi-source measurement values, quality markers, and contextual metadata, serving as authoritative input for subsequent processing. Typical content includes timestamps, device identifiers, sensor identifiers, sampling sequence numbers, measurement values, quality bits, sampling frequency, compartment location, and trigger tags. For example, within the same time window, there might be gearbox vibration envelope statistics, motor phase current transients, battery cell voltage snapshots, and cabin temperature and humidity readings, which are entered into the server as a single message access topic while retaining their original granularity.

[0020] Furthermore, when acquiring sensor data streams from temperature sensors, vibration sensors, current and voltage sensors, pressure sensors, accelerometers, inertial sensors, and environmental sensors, the measured values, timestamps, device identifiers, sensor identifiers, quality markers, and compartment position information are packaged according to a unified data frame specification on the heterogeneous sensor array side. This data is then pushed to the server's access buffer via a fieldbus or publish / subscribe channel, forming a continuous and traceable set of sensor data streams. To ensure consistency in subsequent timing processing, all sensors undergo device identifier and compartment pose binding before factory calibration or deployment. The server performs format verification, quality marker parsing, and packet loss recording on arriving data frames, and generates a temporary index based on arrival order for subsequent cross-source alignment, denoted as the sensor data stream set. For ease of subsequent description, the sensor data stream set can be represented in the following form:

[0021] in, Represents a collection of sensor data streams. This represents the set of sensor indices within a heterogeneous sensor array. Indicates the sensor index. This indicates the data frame number of the sensor. Indicates the sensor-side timestamp. Indicates equipment identification. Indicates sensor identification, Indicates the measured value. Indicates quality marking, This indicates the location information of the compartment; the above indication is used to define the integrity boundary of data items and identification items, which facilitates subsequent preliminary synchronization and timing calibration.

[0022] When performing initial synchronization and timing calibration on the sensor data stream set, the frequency and phase offsets of each sensor clock relative to the reference time are first estimated using the server reference clock or the reference time series synchronized at the edge as a benchmark. Then, affine correction is performed on the timestamp of each data frame to map data from different time domains to a unified time domain. The time affine correction can be expressed as:

[0023] in, This represents the corrected unified timestamp. Indicates sensor Frequency offset correction factor relative to reference time, Indicates sensor Phase offset correction relative to reference time Represents the original timestamp. This is for robust acquisition. and By minimizing the weighted residuals using the set of observable alignment anchors on the reference time axis, a joint estimate of the frequency offset correction coefficient and the phase offset correction amount is obtained:

[0024] in, This represents the set of anchor sample indices used for calibration. Represents sample weights and is derived from quality labels. Representation and Sample The corresponding reference time axis timestamp; by minimizing the weighted residuals, the influence of outliers on the calibration results can be suppressed, ensuring that the frequency offset correction coefficient and phase offset correction remain stable even in the presence of noise and jitter. After completing the time affine correction, the calibration is performed according to a uniform time window length. The corrected data is sliced ​​and aggregated to form a consistent slice set with unified timestamps, equipment identifiers, sensor identifiers, quality markers, and compartment location information. This set is then output as the original multi-source data stream packet, which can be represented as:

[0025] in, Indicates the first A raw multi-source data stream packet, Indicates the length of the time window. This indicates a left-closed, right-open time interval. During this aggregation process, the server performs deduplication and misalignment correction based on the arrival order index and quality marker, ensuring that each original multi-source data stream packet corresponds to the subsequent enhanced sensing stream generation stage. This allows the output of the initial synchronization and timing calibration to be seamlessly transferred to subsequent data gap completion and scene enhancement technologies.

[0026] S120. Based on the original multi-source data stream packets, an enhanced perception stream is generated using data gap completion technology and scene enhancement technology.

[0027] Specifically, data gap completion technology refers to the process of detecting, locating, and traceably filling in time-slice gaps, field gaps, and short-term interruptions in the original multi-source data stream packets caused by sensor jitter, link packet loss, sleep switching, or sampling misalignment. It requires maintaining the continuity of timestamps, the consistency of device identification, and the auditability of quality tags. For example, if exhaust temperature is missing during the 2-second communication jitter caused by wind and waves, data gap completion technology can generate a temperature completion segment with a "completion source" quality tag based on the stable correlation between fuel injection pulse width and turbine speed within the same window and the trend constraints of effective samples before and after, thereby restoring the continuous observation of that time window.

[0028] Scene enhancement technology refers to injecting derived information and virtual quantities related to operating conditions, equipment relationships, and environmental context into the original multi-source data stream packets without changing the traceability of the original observations. This improves the separability and interpretability of subsequent identification of latent couplings. The injected information includes operating condition labels, such as port entry / exit, constant speed cruise, and heavy load acceleration; environmental labels, such as wind and wave levels, sea temperature, and external air pressure; equipment relationship labels, such as mechanical coupling links and electrical coupling links; and verified virtual quantities, such as propeller shaft torque estimation, bus power distribution status, and effective heat exchange indication of cooling circuits, ensuring strict alignment with the original timestamps, equipment identifiers, and compartment positions. For example, under cross-wave conditions, the same time window can be labeled with "cross-wave level 4" and "constant speed cruise," and a virtual quantity with "bus power distribution status" as "increased motor boost ratio" can be generated, enabling subsequent analysis to distinguish between environmental disturbances and equipment disturbances.

[0029] Enhanced perception stream refers to the aligned data stream output after data gap completion and scene enhancement technologies are completed. It contains both real-world sampling dimensions and enhanced dimensions. It retains the original observations and quality labels, and adds operating condition labels, environmental labels, equipment relationship labels, and virtual quantities, forming a unified coded input that can be directly consumed by downstream deduplication, robustness, and semantic segmentation processes. For example, the enhanced perception stream in the same time window carries alignment information such as "gearbox vibration increase", "generator current pulsation intensifies", "bus power distribution status change", "fourth level transverse wave" and "cruise speed", making it easier for subsequent latent layer evolution script comparison to identify the evolutionary order of "mechanical coupling anomaly precedes electrical coupling anomaly".

[0030] Furthermore, based on the original multi-source data stream packets, a data gap completion technique is employed. This technique uses a time-series interpolation algorithm to predict and fill in missing sensor data. Simultaneously, leveraging the correlation between devices, the reasonable values ​​of the lost data are calculated by analyzing the status data of related devices, resulting in a multi-source data stream with data gap completion. In implementation, the missing segments of each sensor channel are first located on a unified time axis, and interpolation estimates are constructed using time-series neighborhood samples. Subsequently, cross-device regression estimates are constructed using related device channels that have mechanical or electrical coupling with the sensor, and adaptive fusion is performed according to confidence weights, thus forming a single and traceable completion result. The specific formula is as follows:

[0031] in, Indicates sensor In time Interpolation estimation, Indicates time The set of neighborhood sample indices Indicates sensor In time Effective observation, The interpolation weights are represented and normalized; the interpolation weights are determined jointly based on temporal proximity and sample quality labels, and can restore continuity in local stationary segments.

[0032]

[0033] in, Indicates sensor In time Cross-device correlation estimation, Indication and Sensors Multi-channel observation vectors with coupling relationships (such as synchronous sampling of motor phase current, gearbox vibration, bus voltage, etc.) Represents the regression coefficient vector. The term represents the bias term; the coefficients are obtained through robust regression of historically aligned segments, which can extrapolate missing values ​​using coupling information when a single channel is unobservable.

[0034]

[0035] in, This represents the unified estimate after completion. This indicates the fusion weights, which are dynamically adjusted based on the interpolation confidence and cross-device regression confidence; when the neighborhood samples are sufficient and of high quality... When the trend is larger, priority is given to using time-local information, and when the consistency of related channels is stronger. The extrapolation is reduced to enhance coupling, and the final output is a multi-source data stream with source and quality tags and data gap filling.

[0036] Based on the original multi-source data stream packets, scene enhancement technology is used to infer equipment status or environmental parameters that cannot be directly measured. A machine learning model is then used to learn and model the relationship between the equipment and the environment, resulting in a scene-enhanced multi-source data stream. During implementation, time-calibrated observation vectors are combined with operating condition context, environmental context, and equipment relationship context to form a feature set. A trained machine learning model infers virtual quantities and labels, including propeller shaft torque indication, bus power distribution status, effective heat transfer indication of cooling circuits, operating condition labels, and environmental labels. Each inference result is appended with a source label and confidence weight. This injects an enhanced dimension highly correlated with coupled identification into the same time window without altering the traceability of the original observations. The specific formula is as follows:

[0037] in, Indicates time The inferred vector of equipment state or environmental parameters that cannot be directly measured. Indicates parameters Machine learning model mapping of representations Represents the original observation vector. A concatenated vector representing the operating condition context, environmental context, and device relationship context; parameters The method was derived from supervised learning based on labeled historical data and determined through cross-validation. Composed of factors such as speed, load command, wind and wave level, seawater temperature, and coupled topology labels, the model uses nonlinear mapping to capture the interaction effects between the equipment and the environment and outputs scene enhancement results with confidence weights.

[0038] The multi-source data streams, complete with data gaps and enhanced with scene-enhanced data streams, are combined to generate an enhanced perception stream containing both real-world sampling and enhancement dimensions. During implementation, a unified time window is used to key-align and merge the two types of results at the field level. A single, uniformly coded record is output for each time window, containing original observations, completed observations, virtual quantities, operating condition labels, environmental labels, equipment relationship labels, and quality markers. The field dictionary is ensured to be consistent with the interface contracts of subsequent deduplication, robustness, and semantic segmentation stages, thus forming a directly consumable enhanced perception stream. To maintain subsequent traceability, source markers and confidence weights are retained for each merged field, and consistency verification values ​​are generated at the time window granularity to support the joint use of multi-source evidence in subsequent latent layer evolution scripts and weak pattern recognition. The specific formula is as follows:

[0039] in, Indicates time Enhanced sensing stream recording, This represents the observation vector completed by filling in data gaps. This represents the virtual variables and label vectors inferred through scene enhancement. The metadata vector represents the data and includes timestamps, device identifiers, sensor identifiers, compartment locations, quality markers, source markers, and confidence weights. The above concatenation is constrained by field consistency and time alignment, thereby seamlessly merging the multi-source data streams that have been filled in with data gaps with the multi-source data streams that have been enhanced by the scene, and outputting an enhanced perception stream that can be directly used for subsequent event-based feature sequence generation.

[0040] S130. The enhanced sensing stream is deredundant, robust, and semantically segmented, and the temporal segments of device fluctuations, energy rheology, and operation actions are extracted to generate event-based feature sequences.

[0041] Specifically, redundancy removal refers to identifying and eliminating repetitive information in the augmented sensing stream that is spatially, channel-wise, or statistically correlated, in order to reduce the dimensionality and noise impact of subsequent processing and retain the observations that are most discriminative of coupling symptoms. For example, redundancy is determined for near-repetitive high-frequency components collected by two adjacent accelerometers of the same gearbox, retaining only the channel with the higher signal-to-noise ratio or its fused principal component, and recording the quality label and source link for the eliminated channel. Robustness refers to anomaly suppression and stabilization processing of the augmented sensing stream in the presence of outliers, impulse noise, and short-term jitter, making statistics and features insensitive to disturbances, while retaining abrupt and gradual changes in the actual operating conditions. For example, if the phase current of a motor spikes under surge electromagnetic interference, robustness marks the spike as an anomaly and uses robust smoothing to recover the continuous trajectory, thus not affecting subsequent energy rheology identification.

[0042] Semantic segmentation refers to dividing a continuous time series into segments with clear semantic boundaries based on equipment operating modes, condition labels, and operational events. Each segment corresponds to interpretable stages such as "start-up," "acceleration," "constant speed," "deceleration," and "shutdown," or operational behaviors such as "valve opening change" and "load command adjustment." For example, based on propulsion commands and speed ramp-up slopes, a single outbound process can be segmented into three segments: "slow-moving," "acceleration disengagement," and "constant-speed cruising," while retaining the start and end times and triggering conditions of each segment. Equipment fluctuations refer to short- or medium-timescale dynamics directly related to changes in the state of mechanical, rotating, or thermal components, including changes in vibration amplitude, harmonic energy, temperature rise, and pressure pulsation. These are used to reveal early clues of abnormal mechanical coupling or component degradation. For example, the appearance of an unconventional octave band in gearbox vibration during load increases is an abnormal pattern of equipment fluctuations.

[0043] Energy rheology refers to the temporal characteristics of energy distribution, transfer, and loss during transmission across components, circuits, and forms (mechanical and electrical work) in a coupled marine propulsion system. It is used to characterize the linkage between electrical and mechanical coupling. For example, during the phase of increasing the proportion of electric motor boost, the bus power distribution shifts from the diesel engine side to the electric motor side, and at the same time, the propulsion shaft torque component structure undergoes a measurable change, which reflects energy rheology. Operational actions refer to control behaviors triggered by the operator or automatic control system and marked with observable data. These include changes in propulsion commands, generator grid connection / disconnection, converter current limiting strategy switching, and cooling circuit valve opening adjustments. Their function is to provide clear causal anchors for data segments. For example, switching from "economy mode" to "performance mode" during automatic cruise is an operational action, triggering linked changes in speed, load, and temperature over the following tens of seconds.

[0044] A temporal segment refers to the smallest continuous time window with a single semantic meaning on a unified time axis after semantic segmentation. It contains a complete set of observational evidence for the same event or operating condition, facilitating statistical analysis and pattern matching at the segment level. For example, a "5-minute window under constant speed cruise (straight waves level 4)" is a temporal segment containing aligned observations and enhancement labels for all relevant channels within that window. An event-based feature sequence encodes multiple temporal segments into a replayable, searchable, and comparable discrete event sequence based on chronological order and causal clues. Each event is accompanied by segment statistical features, context labels, source markers, and quality markers, used for downstream latent layer evolutionary script matching and weak pattern recognition. For example, continuously generating an event-based feature sequence of "acceleration event → torque redistribution event → vibration sideband enhancement event" can provide structured evidence for subsequent output of latent coupling candidate link sets.

[0045] Furthermore, principal component analysis is used to remove redundant signals and irrelevant noise between devices in the enhanced sensing stream, retaining the first key data feature. In implementation, the multi-channel enhanced sensing stream aligned to a unified time axis is represented as a time-sampled vector sequence. Zero-mean and unit-variance standardization is applied to each channel to eliminate dimensional differences. Subsequently, the covariance matrix is ​​calculated and eigenvalue decomposition is performed to obtain the principal axis. The data is then projected onto a principal subspace whose cumulative contribution rate meets a threshold to form the first key data feature. The specific formula is as follows:

[0046] in, For a moment Enhanced sensing stream multi-channel observation vector, This is the mean vector for each channel. Let be the standard deviation vector of each channel. For standardized vectors, Let covariance matrix be the variance matrix. The eigenvector matrix, It is an eigenvalue diagonal matrix. For the first 1 eigenvalue, For the number of channels, To take the front eigenvector submatrices of principal components The first key data feature, For cumulative contribution rate, The cumulative contribution rate is the inter-value; by retaining only the principal components that meet the cumulative contribution rate threshold, the redundancy brought by highly correlated channels is suppressed and the change principal axis most relevant to the coupling symptoms is retained.

[0047] Combining the first key data feature, the enhanced sensing stream is robustly processed by smoothing it using a low-pass filter or a median filter to obtain the second key data feature; during implementation, the first key data feature is... Channel j uses a first-order digital low-pass filter to suppress high-frequency noise, while a windowed median filter is used in channels with impulse interference or spike outliers to preserve edges and steps. The two filters are selected optimally or cascaded to obtain a smooth and high-fidelity feature sequence. The specific formula is as follows:

[0048] in, The second key data feature is in vector form after being smoothed by a low-pass filter. For the smoothing coefficients of the low-pass filter, For the first Channel at time Window length The scalar output after median filter smoothing, median For the median operator, The first key data feature is obtained by applying low-pass and median filters to different noise patterns, respectively, to obtain a second key data feature that is robust to both outlier interference and high-frequency jitter.

[0049] Combining the second key data feature with the device's operating mode and status, an event detection algorithm is used to semantically segment the enhanced sensing stream to obtain the third key data feature. In implementation, operating mode and status labels are introduced as context. An event detection score is constructed based on the deviation between the second key data feature and the mode reference vector. A mode-adaptive threshold is used to generate segmentation boundaries, thereby dividing the continuous time axis into a set of minimal time windows with clear semantics. Each time window is assigned a semantic label and context reference to form the third key data feature. The specific formula is as follows:

[0050] in, For a moment Event detection score, This is the second key data feature. For the running mode label Operation status label The pattern reference vector of the index. For pattern adaptive threshold, For the semantic segmentation boundary set, For semantic time windows, For the first A semantic time window; by detecting deviation events under the mode adaptive threshold, the semantic segmentation is ensured to dynamically adapt to the switching of working conditions and strategies.

[0051] Based on the third key data feature and the enhanced sensing stream, time-series segments related to equipment fluctuations, energy rheology, and operational actions are extracted to generate corresponding event-based feature sequences. During implementation, at the granularity of each semantic time window, statistics related to equipment fluctuations, energy rheology, and operational actions are calculated separately and encapsulated together with the context labels of the time windows as event records. These records form an event-based feature sequence in chronological order, thereby achieving a structured mapping from continuous signals to replayable events. The specific formula is as follows:

[0052] in, For the first A semantic time window, The feature vector is obtained by augmented sensing flow mapping and includes device fluctuation channel, energy rheology channel and operation action channel. In order to be in Fragment-level statistical features within, For event-based feature sequences, and Time windows The running mode label and operation status label, source Source tag, quality For quality labeling; by aggregating three types of evidence in units of semantic time windows and retaining source and quality information, the event-based feature sequence can provide an interpretable, traceable and comparable structured input for the subsequent output of the latent coupling candidate link set through latent layer evolution script and weak pattern recognition algorithm.

[0053] S140. The event-based feature sequence is recombined and compared using the latent layer evolution script and weak pattern recognition algorithm to output a set of latent coupling candidate links.

[0054] Specifically, the latent layer evolution script refers to a set of temporal narrative templates based on historical data, simulation data, and expert knowledge. These templates are used to depict the typical path of a coupled ship propulsion system evolving from a normal state to an abnormal state. The template consists of the event sequence, time window constraints, context labels, spatial mapping, and equipment relationship constraints in the event-based feature sequence, along with triggering conditions and exclusive conditions. For example, in the context of "constant speed cruise, cross wave level 4", the "motor phase current imbalance event" occurs first, followed by the "gearbox vibration side band enhancement event", then the "bus power distribution fluctuation event", and after the "cooling circuit effective heat exchange indicator decline event", a possible abnormal evolution path is formed. This path requires that the events occur in the same propulsion chain topology, adjacent sections, and that the order of events is not reversed.

[0055] Weak pattern recognition algorithms refer to methods for detecting and identifying anomalous patterns with low signal-to-noise ratio, short duration, and low amplitude but consistent characteristics across channels. They emphasize robust extraction from multi-channel, multi-scale evidence in enhanced sensing streams and event-based feature sequences, and use multi-channel consistency constraints, temporal consistency constraints, device coupling consistency constraints, and quality label constraints for filtering and confirmation. For example, under strong environmental disturbances and thermal drift, weak pattern recognition algorithms can extract weak anomalous patterns that are in the same window, topology, and occur synchronously from three types of evidence: "slight upward arching of generator three-phase current harmonic energy," "slight rise in gearbox frequency doubling side band," and "slight deepening of bus voltage transient ripple." These patterns can then be used to support subsequent comparison with the latent layer evolution script.

[0056] Event-based feature sequences refer to discrete event queues assembled from temporal segments obtained after semantic segmentation in chronological order. Each event carries segment statistical features, context labels, spatial mapping, equipment relationships, source markers, and quality markers, covering replayable evidence such as "equipment fluctuation events," "energy rheology events," and "operational action events." For example, a departure process forms an event-based feature sequence of "acceleration event → torque redistribution event → vibration side band enhancement event → cooling fan strategy switching event," which is used to participate in subsequent reassembly and comparison.

[0057] Recombination and comparison refers to splicing, aligning, and verifying the consistency of candidate events in the event-based feature sequence according to the order, time window constraints, spatial mapping, and equipment relationships specified in the latent layer evolution script. Fault-tolerant matching is performed on anchor points that are allowed to be missing, and branches that do not meet the trigger conditions or violate the exclusivity conditions are eliminated. Finally, several matching segments that meet the script constraints and their confidence weights are output. For example, the "motor phase current imbalance event" and the "gearbox vibration side band enhancement event" are aligned in the same propulsion chain topology and the same time window. Then, it is checked whether there is a "bus power distribution fluctuation event" after them. If the order and threshold conditions of the script are met, a valid recombination segment is formed; otherwise, it is marked as inconsistent and discarded.

[0058] The latent coupling candidate link set refers to the set of possible causal propagation paths obtained after reorganization and comparison. Each latent coupling candidate link is encoded by fields such as link segment, triggering condition, evolution priority, spatial mapping, applicable context, and risk level, and retains source and quality tags to support traceability and verification. For example, "Link A: Electrical coupling segment <motor phase current imbalance event> → Mechanical coupling segment <gearbox vibration side band enhancement event> → Energy distribution segment <bus power distribution fluctuation event> → Thermal management segment <cooling circuit effective heat exchange indicator decline event>, triggering condition is 'load step exists and environmental label is transverse wave level 4', evolution priority is high, spatial mapping is 'motor—gearbox—distribution bus—cooling circuit'", this link, together with other latent coupling candidate links in the same batch, is output as a set for priority retrieval and constraint reasoning in fault analysis tree, thereby accelerating root cause localization and improving stability.

[0059] Furthermore, based on historical equipment data, when constructing the latent layer evolution script, the event-based feature sequences covering "normal state to fault state" are first hierarchically clustered according to operating condition labels, environmental labels, and equipment relationship labels. The skeleton of the stable evolution path is extracted and parameterized using transition constraints and time window constraints. The transition probability and residence time distribution parameters of the evolution path are estimated using a joint criterion of maximum likelihood and temporal consistency, resulting in the parameter set that best explains the historical evolution in a given context. This set is used to form an evolution path template that can be used for replay and alignment, as detailed below:

[0060] in, The set of parameters representing the latent layer evolutionary script. This represents the transition probability matrix between events. Indicates the first Adjacent events in a historical sequence The conditional probability of occurrence Indicates the first The dwell time of each evolutionary stage Indicates the first The residence time distribution parameters for each evolutionary stage. This represents the tradeoff coefficient between the transfer term and the time term. Indicates the number of historical sequences. Indicates the first Length of historical sequence This represents the number of stages in the evolutionary path; by solving... The obtained "latent layer evolution script" contains the event sequence, allowed time windows, spatial mapping and triggering exclusive conditions, which can be used as an alignment benchmark under the target conditions.

[0061] A weak pattern recognition algorithm identifies minute fluctuations related to potential equipment failures from event-driven feature sequences and extracts anomalous signals closely associated with failure modes. When weak patterns are obtained, a multi-channel consistency score is first constructed at the segment level for the event-driven feature sequences, followed by robust normalization to suppress occasional impulses and outliers. Then, segments that simultaneously satisfy cross-channel consistency and temporal consistency are selected based on context-adaptive thresholds. Finally, a set of weak patterns with confidence weights and source labels is output, as detailed below:

[0062] in, Indicates the first The original consistency score of each segment, This represents the statistical feature vector of the segment. This represents a mapping that performs scale alignment and inter-channel correlation compression on the feature vectors. This represents the channel weight vector obtained by training on historical annotations.

[0063]

[0064] in, Indicates the first Robust standardized score of each segment, This represents the set of scores for all segments within the same context. This represents the median. Indicates the absolute deviation of the midpoint; when When a fragment exceeds the context adaptation threshold and occurs in a consistent manner with its neighboring fragments in time, the fragment is identified as a "weak pattern" and given a confidence weight obtained during the training phase.

[0065] Weak patterns are recombined and compared with evolutionary paths to generate a set of latent coupling candidate links. When each link includes key stages of fault development, associated events, and temporal sequence, dynamic programming alignment with time window constraints, spatial mapping constraints, and context consistency constraints is employed. This optimally registers the weak pattern sequence with the evolutionary path in the latent layer evolutionary script, and performs fault tolerance processing for missing anchor points and replaceable anchor points. The final output is a list of latent coupling candidate links arranged in evolutionary order with priority and risk level. The alignment objective function consists of a matching score, a skip penalty, and an insertion penalty, as detailed below:

[0066] in, Indicates when the first match is found A faint pattern With evolutionary path number Phase The optimal cumulative score at that time This represents the overall matching score between the two entities in terms of consistency in event type, time window, spatial mapping, and context. This indicates that the penalty for skipping a weak pattern is being skipped. This represents the penalty for inserting a gap in the evolution path. After obtaining the optimal path, the link-level priority and risk level are calculated using the matching pair set and the corresponding score. The link is encoded with "link segment, triggering condition, evolution priority, spatial mapping and applicable context" and aggregated to form a "latent coupling candidate link set". This provides a structured prior for subsequent spatialized coupled light strip scene mapping and fault analysis tree reasoning.

[0067] S150. Map the set of latent coupling candidate links to the three-dimensional hull space and control interface to generate a spatialized coupled light strip scene.

[0068] Specifically, the three-dimensional hull space refers to a spatial representation using a three-dimensional geometric model of the hull as its carrier, encompassing compartment topology, equipment installation points, pipeline routes, and maintenance access routes. It possesses a unified spatial coordinate system, hierarchical structure, and semantic tags, supporting visual queries of equipment pose, compartment boundaries, and maintenance access routes. For example, the engine room deck, port side cabinet area, propulsion bearing housing, and cooling pump assembly have searchable spatial locations and hierarchical paths within the "three-dimensional hull space," facilitating the presentation of events on the correct compartments and equipment entities. The control interface refers to a collection of interactive screens for pilots and maintenance personnel, including a situation overview, equipment panels, alarm lists, timelines, parameter trends, and interactive controls. It supports bidirectional linkage and status synchronization with the "three-dimensional hull space," and provides context menus and playback entry points for each visual element. For example, clicking on the "gearbox component" in the situation overview will link the "three-dimensional hull space" to the gearbox location and load relevant curves for the past ten minutes in the parameter trends, while simultaneously filtering events associated with that equipment in the alarm list.

[0069] The spatialized coupled light strip scenario refers to visualizing the "latently coupled candidate link set" as a continuous light strip with directionality and intensity in the "3D ship space," and simultaneously presenting its temporal evolution and risk level in an interactive scene in the "control interface." The light strip connects the geometric positions of relevant equipment along the link segments, and the width, brightness, and flicker rhythm of the light strip express the propagation intensity, risk level, and activity, respectively. Key light strip nodes are marked with clickable markers to display event timestamps and context labels. For example, the light strip starting from the "motor stator end" extends through the "gearbox housing" to the "power distribution bus," where the "gearbox housing" node is highlighted and labeled "vibration side band enhancement." The overall brightness of the light strip increases in the context of "cross wave level four" to indicate the superposition of environmental disturbances. Key light strip nodes refer to spatial anchor points in the "spatialized coupled light strip scene" that directly correspond to event-based features and can trigger tracing and playback. They carry event time windows, device identifiers, context labels, and quality markers, and support one-click jump to parameter trends and log segments. For example, clicking on the key light strip node of the "bus power distribution fluctuation event" will pop up a sidebar on the control interface, displaying the power ratio curves of motors and diesel engines, alarm entries, and operation action records within the corresponding time window.

[0070] Furthermore, based on the key nodes in each latent coupling link set, the corresponding ship equipment positions and spatial coordinates are determined, and the key nodes, ship equipment positions, and spatial coordinates are mapped into the ship's three-dimensional spatial model. During implementation, based on the unified transformation from equipment pose calibration matrix to hull coordinates to model coordinates, the local coordinates of the key nodes are mapped to three-dimensional spatial model coordinates, while maintaining consistency between the timestamp and equipment identifier, forming a set of spatial anchor points that can be used for geometric rendering, as shown in the following formula:

[0071] in, Indicates the first Homogeneous coordinates of key nodes in the 3D spatial model This represents the local homogeneous coordinates of the critical node on the corresponding device. This represents the equipment pose calibration matrix, which consists of the ship's equipment position and equipment attitude. This represents the scale-offset-alignment matrix that uniformly maps the ship's spatial coordinates to the coordinates of the three-dimensional spatial model. This indicates the device identifier associated with the critical node; calculated across all critical nodes. This yields a set of key node spatial coordinates that are strictly aligned with the three-dimensional spatial model, providing spatial pivot points for the subsequent geometric generation of spatialized light bands.

[0072] Based on the priority, fault risk level, and propagation intensity of each link, different display characteristics are assigned to key nodes and the 3D spatial model to generate spatialized light strips. During implementation, the brightness, width, and flicker frequency of the spatialized light strips are driven by a normalized score based on priority, fault risk level, and propagation intensity. Simultaneously, the hue range is controlled by the fault risk level, and adjacent key nodes are connected by spline curves to generate continuous, renderable geometric paths, as shown in the following formula:

[0073] in, For spatialized light bands in parameters Brightness at that location For spatialized light bands in parameters Width at that point For spatialized light bands in parameters The flashing frequency at that location, For spatialized light bands in parameters Hue at the location; For priority-normalized scoring, To normalize the score for the fault risk level, The score is normalized to reflect the intensity of transmission. These are the lower and upper limits of brightness. These are the lower and upper limits of the spatialized light band width. For the flash frequency range, The hue endpoint from safety to risk; For the center path spline of the spatialized light band, These are control points derived from the spatial coordinates of adjacent key nodes. The basis functions are the cubic splines; through the above attribute mapping and geometric path generation, a continuous, interactive, and readable spatialized light strip is obtained within the three-dimensional spatial model.

[0074] By associating spatialized light strips with the control interface, a spatialized coupled light strip scene is obtained. During implementation, an interface binding relationship and data subscription relationship are established between each key node and each segment of the spatial light strip. This ensures that the timeline, alarm list, and parameter trends of the control interface are synchronized with the spatial light strip in the 3D spatial model. A unified callback is defined for interactive events to achieve source tracing and playback, as shown in the following formula:

[0075] in, To create a list of relationships for the user interface. In the three-dimensional spatial model, the first Individual graphic identifier, For the corresponding interface element identifiers in the control interface (device card, alarm line, time axis mark or parameter curve). The data topic (event stream, state stream, or history replay stream) subscribed to for this element. This serves as a callback pointer for loading the event summary, parameter window, and history fragments after the event is triggered. Number of bound entries; via By establishing a stable mapping between the rendering thread and the UI thread, the interactive state synchronization of the spatially coupled light strip scene can be achieved.

[0076] The spatialized coupled light strip scene is displayed in the control interface. During implementation, the 3D spatial model layer, spatialized light strip layer, and interface overlay layer are rendered synchronously in a layered compositing manner to ensure that the transparency, occlusion relationship, and refresh rhythm of different layers are consistent. The highlighting of key nodes and the visibility of the prompt panel are driven by the timeline, thereby forming a unified situational display that is readable, clickable, and replayable on the control interface. The formula is as follows:

[0077] in, Indicates time The final image frame output to the control interface. This is the rendering result of the three-dimensional spatial model layer. This is the rendering result of the spatialized light band layer. This is the rendering result of the interface overlay, namely the timeline, alarm list, and parameter floating window. To achieve layered transparency coefficients and meet display specifications, This serves as a rendering time index; through this compositing process, the spatially coupled light strip scene is presented on the control interface with a unified visual standard, and maintains frame-level synchronization with the data stream and interaction stream.

[0078] S160. Based on the key light strip nodes in the spatialized coupled light strip scene, generate a root cause analysis report of the coupled ship power system through a fault analysis tree.

[0079] Specifically, key light strip nodes refer to interactive anchor points in a spatially coupled light strip scene that correspond one-to-one with specific event-based features. They carry event time windows, device identifiers, context labels, source markers, and quality markers, and are used to trigger playback, evidence collection, and causal reasoning. For example, the key light strip node for "gearbox vibration side band enhancement" is located on the gearbox shell model. Clicking it will expand the vibration spectrum energy, adjacent device status, and operation action records within that time window.

[0080] Fault analysis trees refer to hierarchical reasoning structures that use causal relationships as the main thread, rules and prior knowledge as constraints, and data evidence as weights. The root node represents the explicit fault symptoms, the intermediate nodes represent the potential mechanism hypotheses, and the leaf nodes represent the verifiable root causes. The optimal explanatory path is obtained through top-down or bottom-up evidence matching and confidence propagation. For example, with "decreased propulsion efficiency" as the root node, it can be decomposed into two major branches: "mechanical coupling anomaly" and "electrical coupling anomaly." Each branch is connected to evidence nodes triggered by key light band nodes, such as "motor phase current imbalance event" and "gearbox vibration side band enhancement event." After reasoning, the most likely causal chain is given.

[0081] Coupled marine propulsion systems refer to a power topology consisting of diesel engines, electric motors, generators, power conversion devices, gearboxes, power distribution buses, batteries, propulsion shafts, and propellers, which simultaneously exhibit mechanical and electrical coupling. Its operational characteristics are manifested in the energy flow across components and the coordinated control of multiple loops; for example, when accelerating out of port, the electric motor assists in increasing thrust, and the mechanical coupling of the diesel engine-gearbox-propulsion shaft system and the electrical coupling of the generator-converter-power distribution bus change synchronously.

[0082] A root cause analysis report is a structured diagnostic document generated based on the results of fault analysis tree reasoning. It includes a fault overview, evolution timeline, optimal causal chain, evidence list, impact assessment, confidence weights, and remedial recommendations. It also retains replayable links and source audit information to facilitate decision-making and closed-loop verification. For example, the report clearly identifies "insufficient lubrication leading to abnormal gear meshing" as the root cause, lists the corresponding key light band nodes, vibration spectrum evidence, and current imbalance evidence, and provides remedial steps such as "checking the lubrication circuit and replacing the filter element" and expected recovery indicators.

[0083] Furthermore, based on the key light strip nodes in the spatially coupled light strip scene, when identifying potential failure events related to equipment failure, evidence scoring and event merging are performed on the key light strip nodes under a unified time axis and three-dimensional ship space coordinates. A set of potential failure events is formed by comprehensively considering temporal consistency, spatial adjacency, contextual consistency, and quality labels. The formula for determining the score from key light strip nodes to potential failure events can be defined as follows:

[0084] in, The overall score for candidate event e. It serves as a measure of temporal consistency, reflecting the proportion of overlap with the time windows of adjacent key optical band nodes; It is a spatial adjacency measure that reflects the normalized negative of the spatial distance to devices within the same propulsion chain topology; It serves as a context consistency measure, reflecting the degree of matching between operating condition labels and environmental labels; The quality label score reflects the source labeling and sampling quality; The weighting coefficients are normalized; when hour( (Using an event determination threshold), the candidate is aggregated into a potential fault event and its time window, device identifier, context label and source label are recorded to form a set of potential fault events for inference.

[0085] Based on the interrelationships between key optical band nodes and equipment, and combining historical operational data with equipment failure modes, a fault analysis tree is constructed. This tree is then used to infer potential failure events. When obtaining causal inference results, a hierarchical structure of "root node—intermediate mechanism node—leaf node" is used to organize the causal chain, transforming equipment failure modes into computable priors and conditional likelihoods. Evidence injection and confidence propagation are then performed on potential failure events. A Naive Bayesian path scoring model can be used to approximate the optimal causal chain, as shown in the following formula:

[0086] in, Mechanism hypothesis In the collection of evidence The posterior probability is given below. Assuming the mechanism is a priori, Potential failure events The conditional likelihood is calculated when the mechanistic hypothesis is true. A joint top-down and bottom-up search is performed on the tree structure to select the mechanistic branch with the largest posterior, and the optimal causal chain and its confidence level, which are consistent with the time order and the topological order of the equipment, are generated as the output of the causal inference result.

[0087] Based on the causal reasoning results, when generating a root cause analysis report that includes the failure event, the causal relationships between key nodes, the root cause of the failure, and the scope of impact, the optimal causal chain and its alternative branches are organized into an evolution timeline according to the strength of evidence and chronological order. A system-level propagation assessment of the impact scope is then conducted to quantify the probability of spillover to relevant equipment and compartments. An impact propagation score can be used to approximate the risk index of the affected component set, as shown in the following formula:

[0088] in, This is a component-level risk index vector. This refers to the component coupling adjacency matrix constructed on the topology of a coupled marine propulsion system. The initial influence vector originating from the root cause node of the optimal causal chain. To truncate the depth for propagation order, The attenuation coefficient is used as the basis for the root cause analysis report, which outputs a fixed structure of "fault overview, evolution timeline, optimal causal chain, evidence list, impact range matrix, confidence weight and treatment recommendations". Each conclusion is accompanied by a source tag and playback link to support auditing and review.

[0089] The user interface provides the root cause analysis report and equipment repair operation guide to the operator so that when performing subsequent equipment repair and troubleshooting, the root cause analysis report items and repair steps can be spatialized and the evidence segments can be played back on a timeline within the user interface.

[0090] Multi-source data is collected via heterogeneous sensor arrays, combined with data gap completion and scene enhancement to ensure data integrity and richness. Then, event-based feature sequences are generated through redundancy removal, robustness enhancement, and semantic segmentation to guarantee the accuracy and stability of feature extraction. Based on this, a set of latent coupling candidate links is output using latent layer evolutionary scripts and weak pattern recognition to reveal potential coupling relationships between devices. Furthermore, the link set is mapped to a spatialized coupled light strip scene, and causal reasoning is performed on key light strip nodes through fault analysis trees to generate root cause analysis reports, thereby achieving system-level health monitoring and risk warning for diesel-electric hybrid power systems. Therefore, this facilitates improved fault diagnosis accuracy for coupled marine power systems.

[0091] This application also provides a coupled marine propulsion system fault diagnosis device, referring to... Figure 2 , Figure 2 This is a schematic diagram of a coupled marine propulsion system fault diagnosis device provided in an embodiment of this application. The device is a server, which includes an acquisition module 21 and a processing module 22. The acquisition module 21 is used to acquire the original multi-source data stream packets sent by the heterogeneous sensor array to the coupled marine propulsion system. The processing module 22 is used to generate an enhanced perception stream based on the original multi-source data stream packets using data gap completion technology and scene enhancement technology. The processing module 22 is also used to perform redundancy removal, robustness, and semantic segmentation on the enhanced perception stream, and extract the temporal segments of equipment fluctuations, energy rheology, and operation actions to generate an event-based feature sequence. The processing module 22 is also used to reconstruct and compare the event-based feature sequence through latent layer evolution scripts and weak pattern recognition algorithms to output a latent coupling candidate link set. The processing module 22 is also used to map the latent coupling candidate link set onto the three-dimensional hull space and control interface to generate a spatialized coupled light strip scene. The processing module 22 is also used to generate a root cause analysis report of the coupled marine propulsion system based on the key light strip nodes in the spatialized coupled light strip scene through a fault analysis tree.

[0092] It should be noted that the above embodiments of the apparatus are only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the apparatus and method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process can be found in the method embodiments, which will not be repeated here.

[0093] This application also provides an electronic device, with reference to... Figure 3 , Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include: at least one processor 31, at least one network interface 34, a user interface 33, a memory 35, and at least one communication bus 32.

[0094] The communication bus 32 is used to enable communication between these components.

[0095] The user interface 33 may include a display screen and a camera. Optionally, the user interface 33 may also include a standard wired interface and a wireless interface.

[0096] The network interface 34 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface).

[0097] The processor 31 may include one or more processing cores. The processor 31 connects to various parts of the server via various interfaces and lines, executing instructions, programs, code sets, or instruction sets stored in the memory 35, and calling data stored in the memory 35 to perform various server functions and process data. Optionally, the processor 31 may be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). The processor 31 may integrate one or a combination of several of the following: Central Processing Unit (CPU), Graphics Processing Unit (GPU), and modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content to be displayed on the screen; and the modem handles wireless communication. It is understood that the modem may also not be integrated into the processor 31 and may be implemented as a separate chip.

[0098] The memory 35 may include random access memory (RAM) or read-only memory. Optionally, the memory 35 may include a non-transitory computer-readable storage medium. The memory 35 can be used to store instructions, programs, code, code sets, or instruction sets. The memory 35 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-described method embodiments, etc.; the data storage area may store data involved in the above-described method embodiments, etc. Optionally, the memory 35 may also be at least one storage device located remotely from the aforementioned processor 31. Figure 3 As shown, the memory 35, which serves as a computer storage medium, may include an operating system, a network communication module, a user interface module, and an application program for a coupled marine power system fault diagnosis method.

[0099] exist Figure 3 In the electronic device shown, the user interface 33 is mainly used to provide an input interface for the user and to obtain the user input data; while the processor 31 can be used to call an application program stored in the memory 35 for a coupled marine power system fault diagnosis method. When executed by one or more processors, the electronic device performs one or more methods as described in the above embodiments.

[0100] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0101] This application also provides a computer-readable storage medium storing instructions. When executed by one or more processors, these instructions cause an electronic device to perform one or more of the methods described in the above embodiments.

[0102] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0103] In the several embodiments provided in this application, it should be understood that the disclosed apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the shown or discussed mutual couplings or direct couplings or communication connections may be through some service interfaces; indirect couplings or communication connections between apparatuses or units may be electrical or other forms.

[0104] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0105] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0106] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned memory includes various media capable of storing program code, such as USB flash drives, portable hard drives, magnetic disks, or optical disks.

[0107] The foregoing description is merely an exemplary embodiment of this disclosure and should not be construed as limiting the scope of this disclosure. Any equivalent changes and modifications made in accordance with the teachings of this disclosure shall still fall within the scope of this disclosure. Those skilled in the art will readily conceive of other embodiments of this disclosure upon considering the specification and the disclosure of practical truth. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not described in this disclosure. The specification and embodiments are considered exemplary only, and the scope and spirit of this disclosure are defined by the claims.

Claims

1. A method for fault diagnosis of a coupled marine propulsion system, characterized in that, The method includes: Acquire raw multi-source data stream packets sent by heterogeneous sensor arrays for coupled marine propulsion systems; Based on the original multi-source data stream packet, an enhanced perception stream is generated using data gap completion technology and scene enhancement technology. The enhanced sensing stream is deredundantized, robustened, and semantically segmented, and time-series segments of device fluctuations, energy rheology, and operational actions are extracted to generate event-based feature sequences. The event-based feature sequence is recombined and compared using latent layer evolution scripts and weak pattern recognition algorithms to output a set of latent coupling candidate links. The set of latent coupling candidate links is mapped onto the three-dimensional ship space and control interface to generate a spatialized coupled light strip scene. Based on the key light band nodes in the spatialized coupled light band scenario, a root cause analysis report of the coupled ship propulsion system is generated through a fault analysis tree.

2. The method for fault diagnosis of coupled marine propulsion systems according to claim 1, characterized in that, The acquisition of the raw multi-source data stream packets sent by the heterogeneous sensor array to the coupled marine propulsion system specifically includes: The heterogeneous sensor array acquires sensor data streams from temperature sensors, vibration sensors, current and voltage sensors, pressure sensors, accelerometers, inertial sensors, and environmental sensors. The sensor data stream is initially synchronized and time-series calibrated to obtain the original multi-source data stream packet.

3. The method for fault diagnosis of coupled marine propulsion systems according to claim 1, characterized in that, The step of generating an enhanced perception stream based on the original multi-source data stream packets using data gap completion technology and scene enhancement technology specifically includes: Based on the original multi-source data stream packet, the data gap completion technology is used to predict and fill in the missing sensor data through time series interpolation algorithm. At the same time, by utilizing the correlation between devices and analyzing the status data of related devices, the reasonable value of the lost data is calculated, and a multi-source data stream with data gap completion is obtained. Based on the original multi-source data stream packet, the device status or environmental parameters that cannot be directly measured are deduced through the scene enhancement technology, and the relationship between the device and the environment is learned and modeled through a machine learning model to obtain the scene-enhanced multi-source data stream. By combining the multi-source data streams that have had their data gaps filled with data with the multi-source data streams that have had their data enhanced by the scene, an enhanced perception stream is generated that includes both the reality sampling dimension and the enhancement dimension.

4. The method for fault diagnosis of coupled marine propulsion systems according to claim 1, characterized in that, The process of deduplicating, robustening, and semantically segmenting the enhanced sensing stream, and extracting temporal fragments of device fluctuations, energy flow, and operational actions to generate an event-based feature sequence specifically includes: Principal component analysis is used to remove redundant signals and irrelevant noise between devices in the enhanced sensing stream, while retaining the first key data features. Combining the first key data feature, the enhanced sensing stream is robustly processed and smoothed by a low-pass filter or a median filter to obtain the second key data feature; Combining the second key data feature, and based on the device's operating mode and status, an event detection algorithm is used to perform semantic segmentation on the enhanced perception stream to obtain the third key data feature; Based on the third key data feature and the enhanced sensing stream, time-series segments related to device fluctuations, energy rheology and operational actions are extracted to generate corresponding event-based feature sequences.

5. The method for fault diagnosis of coupled marine propulsion systems according to claim 1, characterized in that, The step of recombining and comparing the event-based feature sequence with a latent layer evolution script and a weak pattern recognition algorithm to output a set of latent coupling candidate links specifically includes: Based on historical equipment data, the latent layer evolution script is constructed, which describes the evolution path of the equipment from a normal state to a fault state. The weak pattern recognition algorithm identifies minute fluctuations related to potential equipment failures from the event-based feature sequence and extracts abnormal signals closely related to the failure mode to obtain the weak pattern. The weak patterns are recombined and compared with the evolution path to generate the set of latent coupling candidate links, wherein each link contains the key stages of fault development, related events, and time sequence.

6. The method for fault diagnosis of coupled marine propulsion systems according to claim 1, characterized in that, The step of mapping the latent coupling candidate link set onto the three-dimensional ship space and control interface to generate a spatialized coupled light strip scene specifically includes: Based on the key nodes in each of the aforementioned latent coupling links, the corresponding ship equipment locations and spatial coordinates are determined, and the key nodes, ship equipment locations, and spatial coordinates are mapped into the ship's three-dimensional spatial model. Based on the priority, fault risk level, and propagation intensity of each link, different display characteristics are assigned to the key nodes and the three-dimensional spatial model to generate spatialized light strips; By associating the spatialized light strip with the control interface, the spatialized coupled light strip scene is obtained; The spatialized coupled light strip scene is displayed in the control interface.

7. The method for fault diagnosis of coupled marine propulsion systems according to claim 1, characterized in that, The step of generating a root cause analysis report for the coupled ship propulsion system based on key light band nodes in the spatialized coupled light band scene, using a fault analysis tree, specifically includes: Based on the key light strip nodes in the spatialized coupled light strip scene, identify potential fault events related to equipment failure; Based on the relationship between the key optical band nodes and the equipment, combined with historical operating data and equipment failure modes, a fault analysis tree is constructed, and the fault analysis tree is used to reason about the potential failure events to obtain causal reasoning results. Based on the causal reasoning results, a root cause analysis report is generated, which includes the failure event, the causal relationship between key nodes, the root cause of the failure, and the scope of its impact. The root cause analysis report and equipment repair operation guide are provided to the operator through the control interface to facilitate subsequent equipment repair and troubleshooting.

8. A coupled marine propulsion system fault diagnosis device, characterized in that, The apparatus is used to perform the coupled marine propulsion system fault diagnosis method as described in any one of claims 1 to 7, the apparatus comprising an acquisition module and a processing module, wherein... The acquisition module is used to acquire the original multi-source data stream packets sent by the heterogeneous sensor array to the coupled ship propulsion system; The processing module is used to generate an enhanced perception stream based on the original multi-source data stream packet using data gap completion technology and scene enhancement technology; The processing module is also used to perform redundancy removal, robustness enhancement, and semantic segmentation on the enhanced sensing stream, and extract time-series segments of device fluctuations, energy rheology, and operational actions to generate event-based feature sequences. The processing module is also used to reconstruct and compare the event-based feature sequence using a latent layer evolution script and a weak pattern recognition algorithm, and output a set of latent coupling candidate links. The processing module is also used to map the latent coupling candidate link set to the three-dimensional hull space and the control interface to generate a spatialized coupling light strip scene. The processing module is also used to generate a root cause analysis report of the coupled ship propulsion system based on the key light band nodes in the spatialized coupled light band scene through a fault analysis tree.

9. An electronic device, characterized in that, The electronic device includes a processor, a memory, a user interface, and a network interface. The memory is used to store instructions. The user interface and the network interface are both used to communicate with other devices. The processor is used to execute the instructions stored in the memory to cause the electronic device to perform the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions that, when executed, perform the method as described in any one of claims 1 to 7.

Images