Method and system for detecting abnormity of air cooler of marine main engine

Abstract

The invention discloses a ship main engine air cooler anomaly detection method and system, and relates to the field of ship equipment detection, and the method comprises the steps: obtaining historical operation data when an air cooler is in a healthy state; preprocessing the historical operation data to obtain preprocessed historical operation data; constructing an adaptive resonance theory network based on the preprocessed historical operation data, and repeatedly training until a health clustering category set representing a health state is obtained; performing same preprocessing on real-time operation data collected in real time, and inputting the preprocessed real-time operation data into the trained adaptive resonance theory network to judge whether the real-time operation data belongs to any health clustering category in the health clustering category set or not; and when it is judged that the real-time operation data does not belong to any healthy clustering category, it is judged that the air cooler operates abnormally, and an alarm is triggered. The method has the effect of adapting to complex ship operation conditions.

Description

Technical Field

[0001] This application relates to the field of ship equipment testing, and in particular to a method and system for detecting abnormalities in ship main engine air coolers. Background Technology

[0002] As a crucial component of a ship's propulsion system, the air cooler's primary function is to cool the air entering the main engine, ensuring its operation within a suitable temperature range. This is vital for the safe and stable navigation of the ship. However, due to the complex and variable operating environment of ships, air coolers are subjected to harsh conditions such as high temperature, high humidity, and high salt spray, making them prone to abnormalities such as blockage, leakage, and decreased heat transfer efficiency. Currently, traditional methods for detecting abnormalities in ship air coolers mainly rely on periodic manual inspections and traditional detection methods based on fixed thresholds. Manual inspections suffer from long inspection cycles, high labor intensity, and strong subjectivity, making it difficult to capture abnormal changes in the air cooler in real time. Detection methods based on fixed thresholds cannot adapt to the dynamic changes in ship operating conditions. When a ship operates under different speeds, loads, and ambient temperatures, the normal operating data range of the air cooler will change accordingly. Fixed thresholds are prone to false alarms or missed alarms, failing to meet the needs of practical applications. Therefore, developing a method and system that can adapt to changes in ship operating conditions and accurately detect abnormalities in ship air coolers in real time is of significant practical importance. Summary of the Invention

[0003] To improve upon the high error rate and low adaptability of traditional detection methods, this application provides a method and system for detecting abnormalities in marine main engine air coolers.

[0004] Firstly, the method for detecting abnormalities in a marine main engine air cooler provided in this application adopts the following technical solution: S1, acquiring historical operating data of the air cooler when it is in a healthy state; S2. Preprocess the historical operation data to obtain preprocessed historical operation data; S3. Construct an adaptive resonance theory network based on preprocessed historical running data, and repeat training until a set of healthy cluster categories representing the health status is obtained. S4. After performing the same preprocessing on the real-time collected running data, input it into the trained adaptive resonance theory network to determine whether the real-time running data belongs to any one of the healthy cluster categories in the healthy cluster category set. S5. When it is determined that the real-time operating data does not belong to any health cluster category, the air cooler is deemed to be operating abnormally and an alarm is triggered.

[0005] By adopting the above technical solution, historical operating data on the health status of the air cooler is acquired and preprocessed. An adaptive resonance theory network is then constructed and trained to form a set of health cluster categories representing the health status. Real-time operating data, after the same preprocessing, is input into this network and matched with the corresponding health cluster category. If the current operating data does not belong to any health cluster category, the air cooler is deemed to be operating abnormally, and an alarm is triggered. This solution automatically establishes health benchmarks using historical data, simplifying the detection process and eliminating the need for additional hardware or manual intervention, significantly reducing maintenance costs and operational burden. Simultaneously, cluster-based difference analysis improves the accuracy and real-time performance of anomaly detection, enabling timely warnings of potential faults, thereby enhancing the reliability and safety of air cooler operation and achieving efficient and intelligent operation and maintenance management.

[0006] Optionally, the preprocessing of historical operating data includes data cleaning of the historical operating data, removing outliers and noisy data, and then normalizing the data.

[0007] By employing the aforementioned technical solution, historical operational data is cleaned to remove outliers and noise, ensuring the purity and reliability of the input historical operational data and effectively eliminating the potential interference of invalid data to subsequent analysis. Next, normalization processing is performed to unify historical operational data with different dimensions to a standard range, resolving the issue of feature scale differences and making it easier for the adaptive resonance theory network to converge and stabilize clustering results during training. This preprocessing step significantly improves the consistency and quality of historical operational data, providing solid data support for the establishment of healthy cluster categories, thereby enhancing the accuracy of subsequent real-time operational data analysis. Furthermore, through an automated preprocessing workflow, large-scale historical operational data can be processed efficiently without manual intervention, reducing computational complexity, minimizing error accumulation, and ultimately improving the robustness of air cooler anomaly detection and alarm response speed.

[0008] Optionally, S3 includes constructing an adaptive resonance theory network; inputting each preprocessed historical running data into the adaptive resonance theory network, calculating the similarity between the input historical running data and existing healthy cluster categories; if the similarity with any of the healthy cluster categories reaches a warning parameter, then the historical running data is assigned to that healthy cluster category and the category weight is updated; if the similarity with all healthy cluster categories does not reach the warning parameter, then a new healthy cluster category is created and the current historical running data sample features are stored; iterating through all historical running data until the adaptive resonance theory network converges, forming a set of healthy cluster categories representing the health status.

[0009] By employing the aforementioned technical solution, an adaptive resonance theory network is constructed. Preprocessed historical operational data is sequentially input, and its similarity to existing healthy cluster categories is calculated. When the similarity reaches a warning parameter, the preprocessed historical operational data is assigned to the corresponding category, and the weights are dynamically updated; otherwise, a new healthy cluster category is created to store the features of the historical operational data. All historical operational data is iteratively processed until the network converges, ultimately forming a stable set of healthy cluster categories representing the health status. This solution achieves automated learning and construction of health benchmarks without manual annotation or external intervention, significantly simplifying the training process of the adaptive resonance theory network model. The dynamic strategy based on the warning parameter enhances the adaptability and robustness of the adaptive resonance theory network model, flexibly handling data fluctuations and new data samples, avoiding the omission of potential health patterns. Simultaneously, through iterative updates and weight optimization, a high-quality cluster set is efficiently converged, improving the accuracy and reliability of health status representation, providing a solid data foundation for subsequent anomaly detection, thereby reducing maintenance costs and enhancing the intelligence level of the air cooler detection system.

[0010] Optionally, the warning parameters are dynamically adjusted based on navigation environment parameters, specifically including real-time acquisition of wind speed and wave height of the ship's environment; dynamic calculation of warning parameters based on changes in wind speed and wave height; matching the corresponding learning rate based on the dynamic numerical range of the warning parameters; and inputting the warning parameters and learning rate into the adaptive resonance theory network to update the network's healthy clustering category generation and weight adjustment logic.

[0011] By employing the above technical solution, wind speed and wave height data of the ship's navigation environment are acquired in real time, the range of warning parameters is dynamically calculated, and a corresponding learning rate is matched accordingly. The dynamic warning parameters and learning rate are input into an adaptive resonance theory network, which updates the network's generation logic and weight adjustment mechanism for healthy cluster categories in real time. This solution significantly improves the environmental adaptability of the adaptive resonance theory network model, enabling the warning parameter thresholds to autonomously adjust with fluctuations in seawater conditions, avoiding misjudgments or missed detections caused by fixed parameters in complex environments. The dynamic learning rate matching mechanism enables fine-grained control of model sensitivity, enhancing fault tolerance in harsh environments and improving discrimination accuracy in stable environments. By synchronizing environmental parameters with the network learning strategy in real time, the timeliness and reliability of healthy cluster representation are ensured, significantly improving the robustness and early warning accuracy of the anomaly detection system in real navigation scenarios, thereby optimizing the operational efficiency of ship equipment and reducing the risk of sudden failures.

[0012] Optionally, the learning rate corresponding to the dynamic numerical interval matching based on the warning parameter includes defining the upper limit and lower limit of the critical interval of the warning parameter using a piecewise function strategy; setting the learning rate to a constant high value when the warning parameter is less than the lower limit of the critical interval; setting the learning rate to a constant low value when the warning parameter is greater than the upper limit of the critical interval; and linearly decreasing the learning rate as the warning parameter increases when the dynamic warning parameter is between the lower limit and the upper limit of the critical interval.

[0013] By adopting the above technical solution, a critical range and upper and lower limits for the warning parameter are defined. When the warning parameter is below the lower limit, a constant high learning rate is set; when it is above the upper limit, a constant low learning rate is set. Within the critical range, the learning rate decreases linearly with the increase of the warning parameter. This dynamic mapping mechanism significantly optimizes the model's environmental responsiveness. A constant high learning rate ensures rapid detection of subtle anomalies in low-interference environments (i.e., low warning parameters), while a constant low learning rate suppresses false alarms in strong-interference environments (i.e., high warning parameters). The linear decreasing strategy achieves smooth sensitivity adjustment in the transition range, avoiding model oscillations caused by abrupt changes in the learning rate. By accurately matching environmental fluctuations with model learning intensity, the system ensures detection stability under harsh sea conditions and improves discrimination accuracy under stable operating conditions, significantly enhancing the robustness and adaptability of the anomaly detection system, thereby reducing ship maintenance costs and improving fault prevention efficiency.

[0014] Optionally, determining whether real-time running data belongs to any health cluster category includes calculating the difference between the real-time running data and each health cluster category in the health cluster category set; wherein, the difference is the Euclidean distance between the real-time running data and the feature centers of each health cluster category in the health cluster category set; when the difference between the real-time running data and all health cluster categories in the health cluster category set exceeds a set threshold, it is determined that the real-time running data does not belong to any health cluster category; when the difference between the real-time running data and any health cluster in the health cluster category set is less than or equal to the set threshold, it is determined that the real-time running data belongs to that health cluster category.

[0015] By employing the aforementioned technical solution, the Euclidean distance between the real-time operating data and the feature centers of each health cluster in the health cluster category set is calculated and used as the dissimilarity score. When all dissimilarity scores exceed a set threshold, the real-time operating data is determined not to belong to any health cluster category; when any dissimilarity score does not have a set threshold, the real-time operating data is determined to belong to any health cluster category. This solution replaces subjective experience-based judgment with mathematical modeling, significantly improving the accuracy of status identification. The low complexity of Euclidean distance calculation ensures real-time response capability, meeting the high-frequency monitoring needs of industrial scenarios. Simultaneously, the quantified dissimilarity score provides traceable data support for early fault diagnosis, strengthening the system's ability to detect latent performance degradation, ultimately achieving full-dimensional intelligent perception and risk warning closed-loop management of the air cooler's health status.

[0016] Optionally, it also includes step S6: establishing a multi-level response mechanism based on the difference value range; when the difference value is in the first range, triggering an early warning signal and starting the ship monitoring system data recording function; when the difference value is in the second range, triggering an emergency alarm signal and activating automatic adjustment of the main engine load; when the difference value is in the third range, triggering a shutdown protection signal and cutting off the main engine air supply circuit; wherein the boundary values ​​of the first range, the second range and the third range are dynamically adjusted.

[0017] By adopting the above technical solution, a multi-level response mechanism based on the difference value range is established. This step realizes the graded handling of abnormal conditions. Early warning signals, emergency alarm signals, or shutdown protection signals are triggered according to the difference value falling within different ranges, and the ship's monitoring system data recording function is activated accordingly, the main engine load is automatically adjusted, or the main engine air supply circuit is cut off. Simultaneously, the boundary values ​​of the first, second, and third ranges are dynamically adjusted, enabling the system to adapt to changes in the operating environment, significantly improving the flexibility and accuracy of the response mechanism. This design avoids the risk of false alarms or missed alarms caused by a single threshold, optimizes resource allocation, records only data to support post-event analysis in cases of minor anomalies, adjusts the load in a timely manner to prevent the problem from worsening in cases of moderate anomalies, and forces a shutdown to protect equipment safety in cases of severe anomalies. The multi-level response combined with dynamic boundary values ​​reduces the need for human intervention, enhances the adaptability and real-time performance of the air cooler monitoring system, and effectively prevents major failures.

[0018] Optionally, the dynamic adjustment of the boundary values ​​of the first interval, the second interval, and the third interval includes periodically calling the set of differences consisting of differences; calculating the distribution density function of the set of differences; and adjusting the boundary values ​​of each interval according to the peak offset direction and magnitude of the distribution density function.

[0019] By employing the aforementioned technical solution, the system periodically calls upon a set of differences comprised of differences and calculates its distribution density function. Based on the peak offset direction and amplitude, the boundary values ​​of the first, second, and third intervals are dynamically adjusted. This step achieves intelligent adaptive optimization of the boundary values, significantly improving the accuracy and adaptability of the multi-level response mechanism. Based on the analysis of the distribution density function, the system can automatically capture the overall trend of real-time operational data changes and optimize interval boundaries in real time, effectively reducing threshold failure issues caused by environmental fluctuations or equipment aging. This dynamic adjustment process enhances the sensitivity and robustness of anomaly detection, avoids the risk of false alarms or missed alarms that may be caused by fixed boundary values, and ensures that the optimal alarm response strategy can be matched under different operating conditions.

[0020] Secondly, the system for detecting abnormalities in a ship's main engine air cooler provided in this application includes a data acquisition module, a data preprocessing module, a training module, and an abnormality detection module; The data acquisition module is used to collect historical and real-time operating data of the ship's main engine air cooler. The data preprocessing module, connected to the data acquisition module, is used to preprocess historical and real-time operational data. The training module, connected to the data preprocessing module, is used to determine the number of input layer nodes of the adaptive resonance theory network, construct the adaptive resonance theory network structure, and train the network based on the preprocessed historical running data to form a set of healthy cluster categories representing the health status. The anomaly detection module connects the data preprocessing module and the training module. It is used to input the preprocessed real-time running data into the trained adaptive resonance theory network and determine whether the real-time running data belongs to any healthy cluster category in the healthy cluster category set.

[0021] Understandably, the system for detecting abnormalities in the ship's main engine air cooler provided in the second aspect above is used to execute the method provided in this application. Therefore, the beneficial effects it can achieve can be referred to the beneficial effects in the corresponding method, and will not be repeated here.

[0022] In summary, this application includes at least one of the following beneficial technical effects: 1. By constructing an adaptive resonance theory network and training it based on preprocessed historical operating data to generate a set of healthy cluster categories representing the health status as a benchmark, and combining this with real-time operating data to calculate the difference, the system determines whether a matching healthy cluster category can be obtained from the real-time operating data, thus achieving automated real-time detection of the air cooler's operating status. Utilizing Euclidean distance to quantify the difference significantly improves the sensitivity and accuracy of anomaly detection while avoiding manual intervention. This effectively solves the problems of traditional methods relying on fixed thresholds and having high false alarm rates, thereby reducing equipment failure risks and maintenance costs.

[0023] 2. Based on the numerical range of the difference degree, the system divides the system into three intervals: the first interval, the second interval, and the third interval. A graded response strategy is designed, which triggers data recording for early warning signals, activates host load adjustment for emergency alarms, and cuts off the gas supply circuit for shutdown protection. By periodically analyzing the peak shift of the distribution density function of the difference degree set, the system dynamically adjusts the interval boundary values, enabling the response mechanism to adapt to changes in equipment operating status and significantly improving alarm accuracy. Combined with real-time detection of the host load change rate after an emergency alarm and an automatic alarm level escalation mechanism, a closed-loop optimization system for abnormal response is formed, effectively solving the problem of response delay in sudden operating conditions, minimizing cascading damage to equipment, and ensuring navigation safety.

[0024] 3. From data acquisition, preprocessing, network training to anomaly detection, each module is seamlessly integrated through standardized processes. The data preprocessing module ensures input quality through cleaning and normalization; the training module adaptively matches the number of nodes in the input layer to stabilize clustering results; and the anomaly detection module accurately derives judgment results and triggers corresponding alarm signals promptly and accurately. This system-level design significantly enhances the robustness and real-time performance of anomaly detection, solves the problems of data silos and response disconnect in traditional solutions, and achieves end-to-end automated management from state awareness to proactive protection, effectively extending equipment life and optimizing ship energy efficiency control. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 This is a flowchart illustrating the method for detecting abnormalities in the air cooler of a marine main engine provided in Embodiment 1 of this application; Figure 2 This is a schematic diagram of the adaptive resonance theory network for the method of detecting abnormalities in the air cooler of a ship's main engine provided in Embodiment 1 of this application; Figure 3 This is a schematic diagram of the system for detecting abnormalities in the air cooler of a ship's main engine, provided in Embodiment 1 of this application; Figure 4 This is another flowchart illustrating the method for detecting abnormalities in the air cooler of a ship's main engine provided in Embodiment 2 of this application; Figure 5 This is another structural schematic diagram of the system for detecting abnormalities in the air cooler of a ship's main engine provided in Embodiment 2 of this application. Detailed Implementation

[0027] This application discloses a method and system for detecting abnormalities in a marine main engine air cooler, which is described below in conjunction with the appendix. Figure 1-5 This application will be described in further detail.

[0028] Reference Figure 1-3 A method for detecting abnormalities in a ship's main engine air cooler includes S1, acquiring historical operating data of the air cooler when it is in a healthy state.

[0029] Historical operating data refers to the set of operating data continuously collected by multiple sensors installed on the ship's main engine related systems when the ship's air cooler is in a healthy state and operating normally. The operating data includes main engine load, air cooler air inlet temperature, air cooler air outlet temperature, air cooler air inlet pressure, air cooler air outlet pressure, cooling water flow rate, cooling water inlet temperature, cooling water outlet temperature, cooling water inlet pressure, and cooling water outlet pressure.

[0030] Specifically, by deploying sensors at key locations on the main engine and air cooler during ship navigation, the system detects and records the aforementioned operational data in real time. This ensures that the data covers the stable operating cycle of the air cooler under typical operating conditions. The collected raw historical operational data is directly transmitted to a historical database for storage, forming a benchmark historical operational data set for subsequent network training. During the data acquisition phase, the system only performs data collection on normal, healthy operating conditions to avoid introducing abnormal interference. Furthermore, the acquisition process utilizes the sensor's built-in calibration mechanism to ensure data consistency and reliability, enabling the accumulation and archiving of historical data without manual intervention.

[0031] S2. Preprocess the historical operation data to obtain preprocessed historical operation data.

[0032] The preprocessing process comprises two key stages: data cleaning and normalization. These stages aim to eliminate interference factors in the original historical operating data and standardize its scale. Data cleaning uses automatic algorithms to identify and remove outliers and noisy data from the collected historical operating data, ensuring that the dataset contains only valid information about the air cooler's normal operating condition. Normalization employs scaling algorithms to transform operating data with different dimensions into a unified numerical range, eliminating the impact of dimensional differences between variables on subsequent model training, ultimately generating standardized preprocessed historical operating data. Specifically, the normalization process uses a max-min normalization algorithm to transform historical operating data with different dimensions into a unified numerical range of zero to one. The original value of each historical operating data point is subtracted from its minimum value in the historical operating data set, and then divided by the difference between its maximum and minimum values. This algorithm is automatically applied to all historical operating data, ensuring consistent scale without manual intervention. This historical operating data can be directly input into the adaptive resonance theory network for training. During preprocessing, no manual intervention is required; the cleaning and normalization operations are automatically performed through built-in logic, maintaining the consistency and reliability of the historical operating data.

[0033] S3. Construct an adaptive resonance theory network based on preprocessed historical operating data, and repeat training until a set of healthy cluster categories representing the health status is obtained.

[0034] Among them, reference Figure 2 The Adaptive Resonance Theory Network (ARN) is a neural network architecture with real-time unsupervised learning capabilities. Its core mechanism lies in dynamically adjusting its internal clustering structure to adapt to changes in the feature distribution of input data while maintaining stability of learned patterns. The network consists of a two-layer competitive structure: a comparison layer and a recognition layer. The comparison layer receives the input vector and matches it with the cluster prototypes (healthy cluster categories) stored in the recognition layer. The recognition layer controls the generation of new pattern categories based on a warning data threshold. When the similarity between the input real-time operating data and an existing healthy cluster category is higher than the warning parameter, the network updates the weight of that healthy cluster category to absorb new features; if the similarity is lower than the warning parameter, a new healthy cluster category is created. In the application of anomaly detection in ship air coolers, this network dynamically generates a set of healthy cluster categories by continuously learning from normal operating data. It does not require pre-defining the number of categories or abnormal samples. Its incremental learning characteristic can adapt to the changing operating environment of ships, significantly different from traditional static clustering algorithms. A healthy cluster category refers to the set of feature patterns formed by the ARN analyzing normal operating data of the air cooler; each category represents a health status benchmark under a specific operating condition. The warning parameter is a similarity threshold that determines whether to create a new category; its value dynamically reflects changes in the navigation environment. The learning rate controls the sensitivity of network weight updates.

[0035] Specifically, the number of input layer nodes in the adaptive resonance theory network is determined based on the data dimension of historical operating data. Each input node's neuron corresponds to one piece of historical operating data. For example, in this embodiment, the historical operating data includes historical host load, historical air cooler inlet temperature, historical air cooler outlet temperature, historical air cooler inlet pressure, historical air cooler outlet pressure, historical cooling water flow rate, historical cooling water inlet temperature, historical cooling water outlet temperature, historical cooling water inlet pressure, and historical cooling water outlet pressure—a total of ten historical operating data points. Therefore, the corresponding number of input layer neurons is set to 10. Based on the number of neurons in the input layer nodes, an adaptive resonance theory network topology is constructed. This network topology includes three logical components: an input layer, a comparison layer, and a recognition layer. The input layer receives standardized data vectors; the comparison layer matches the input data with the healthy cluster categories stored in the recognition layer through similarity calculation; and the recognition layer activates the most matching healthy cluster category through a competition mechanism. The network initialization phase includes assigning random initial values ​​to the weight matrix from the comparison layer to the recognition layer, setting warning parameters to control the sensitivity of new cluster category generation, and configuring learning rate data to adjust the cluster prototype update speed, thus establishing a basic framework for subsequent unsupervised training. Then, preprocessed historical running data is input into the network one by one. The similarity between each historical running data point and existing healthy cluster categories is calculated. When a single healthy cluster category meets the warning parameter threshold, the historical running data is assigned to that healthy cluster category and its weight vector is updated. If multiple healthy cluster categories simultaneously reach the warning parameter, the category with the highest similarity value is selected for merging. When the similarity of all existing categories is below the warning parameter, a new healthy cluster category is created and the current historical running data features are stored. During this process, real-time wind speed and wave height data can be obtained from the ship's weather station. The warning parameter is dynamically calculated based on the real-time acquired navigation environment parameters such as wind speed and wave height, and the calculation formula is as follows: Warning Parameter ,in, These are the basic warning parameters, typically set manually. They are usually used in general vessel applications requiring clustering categories, where a balance needs to be struck between sensitivity and false alarm rate. A value between 0.3 and 0.6 is acceptable. Increased wind speed or wave height indicates significant changes in sea conditions. Increase and improve the classification and clustering standards to avoid false vibration alarms caused by changes in sea state. The environmental sensitivity coefficient is obtained through training with historical operational data. , These are the weighted parameters for wind speed and wave height, respectively.

[0036] The system employs a piecewise function strategy to match the learning rate based on the numerical range of the dynamic warning parameter. Specifically, it defines an upper and lower limit for the critical interval of the warning parameter. When the warning parameter is below the lower limit, the learning rate is set to a constant high value to accelerate weight updates. When the warning parameter is above the upper limit, the learning rate is set to a constant low value to stabilize the network. When the warning parameter is between the lower and upper limits, the learning rate decreases linearly as the warning parameter increases, ensuring the network adapts to environmental changes. The training process continuously iterates through historical data samples, updating the class weights through dynamically adjusted learning rates until the network converges and no longer generates new healthy clusters (i.e., the cluster categories no longer change significantly), forming a stable set of healthy cluster categories.

[0037] This dynamic adjustment mechanism for warning parameters directly responds to the uncertainties of the navigation environment. Wind speed and wave height, as key environmental variables, directly affect the heat exchange efficiency and parameter fluctuations of the air cooler. Under high sea state conditions, operating parameters naturally exhibit greater dispersion. In this case, lowering the warning parameter threshold can avoid misjudging normal fluctuations, while in calm sea states, it is necessary to increase detection sensitivity. The dynamic matching strategy of the learning rate aims to resolve the contradiction between environmental changes and network stability. When the warning parameter decreases (corresponding to a harsh environment), a high learning rate is set to accelerate the network's rapid learning of new operating conditions and the formation of new healthy cluster categories. When the warning parameter increases (corresponding to a stable environment), a low learning rate is used to prevent existing cluster weights from being interfered with by noise. The design of linearly decreasing learning rate when the warning parameter is in the transition range ensures that the network parameter adjustment evolves continuously and smoothly with environmental changes. The rule of selectively merging multiple matching healthy cluster categories strengthens the determinism of cluster boundaries and avoids the input historical operating parameter data being incorrectly assigned to secondary categories. The entire iterative process continuously optimizes the network structure through real-time environmental perception, so that the final healthy cluster category set has both environmental adaptability and state representation accuracy, breaking through the technical bottleneck of high false alarm rate of traditional fixed parameter models when sea state changes.

[0038] It should be noted that the core operating parameters in this method include air inlet and outlet temperatures, cooling water inlet and outlet temperatures, and cooling water flow rate. When the ship is in stable navigation conditions and the cooling system structure remains unchanged, the main engine load and pressure parameters can be selectively omitted, significantly improving the system's real-time performance while ensuring anomaly detection accuracy. This is particularly suitable for shipborne embedded platforms with limited computing resources. The specific parameter simplification strategy needs to be dynamically configured based on the actual ship's sensor accuracy, sea conditions, and equipment aging. However, in cases of sudden environmental changes, equipment aging, system modifications, or other abnormal changes, all operating parameters must be used. Furthermore, this method must specifically consider the impact of the seawater environment. The continuous effect of seawater salinity can cause progressive scaling and deposition on the heat exchange surface of the air cooler, leading to a monthly decrease in heat transfer efficiency. This slow performance degradation can easily be misjudged as a sudden failure. To address this issue, the dynamic pattern expansion capability of the ART network can be utilized. When a continuous 30-day monotonous upward trend in cooling water outlet temperature is detected while other parameters remain stable, a new category of high-salinity operating mode can be automatically generated and included in the healthy cluster category set, avoiding false alarms and providing maintenance cycle predictions. To address the drift in detection benchmarks caused by regional differences in seawater temperature, this method adds a seawater temperature sensor below the ship's waterline to collect real-time data. A temperature correction formula with a sea area characteristic coefficient is used to compensate for the cooling water temperature parameter. Simultaneously, seawater temperature is input as another operational parameter feature into the network to construct an environmental adaptability model, reducing the impact of seawater temperature variations in different regions. When the ship encounters a combined environment of high salinity and high temperature seawater, the system automatically triggers a dual verification mechanism. By comparing historical operational data from the same sea area and temporarily adjusting the anomaly detection threshold, the system ensures detection reliability under extreme conditions such as storms.

[0039] S4. After performing the same preprocessing on the real-time collected running data, input it into the trained adaptive resonance theory network to determine whether the real-time running data belongs to any healthy cluster category in the healthy cluster category set.

[0040] Real-time operational data refers to the set of air-cooler operating parameters collected in real-time by ship sensors, with data dimensions completely consistent with historical operational data. The same preprocessing refers to performing data cleaning, outlier removal, and normalization operations on the real-time data in the same manner as on the historical data. The trained adaptive resonance theory network refers to a network model that has completed health state clustering learning, internally forming a stable set of healthy cluster categories representing the normal operating modes of the air-cooler. The healthy cluster category set refers to the set of cluster categories representing all known healthy operating modes of the air-cooler, learned and solidified by the adaptive resonance theory network based on historical health operational data during the training phase. Determining whether a data belongs to any healthy cluster category in the healthy cluster category set involves calculating the difference between the real-time operational data vector and the feature center point of each healthy cluster category in the healthy cluster category set, and making a matching decision based on a preset difference threshold.

[0041] Specifically, the system continuously acquires real-time operating parameters of the ship's main engine air cooler, including main engine load, air inlet and outlet temperatures and pressures, and cooling water flow rate and inlet and outlet temperatures and pressures. These real-time operating data streams undergo preprocessing operations consistent with historical data, including data cleaning to remove outliers and noise interference, and normalization to ensure data scale uniformity, generating standardized real-time operating data vectors. These vectors are then input into a trained adaptive resonance theory network. A permanently stored set of healthy cluster categories is then invoked, containing multiple healthy cluster categories and their corresponding feature centroid vectors. For each input real-time operating data vector, the Euclidean distance between it and the feature centroid of each healthy cluster category in the set is calculated. This Euclidean distance represents the degree of deviation between the current state and a specific health pattern. The system then compares the calculated differences between the real-time operating data vector and the feature centroids of all healthy cluster categories with a preset difference threshold. If at least one healthy cluster category exists, and its corresponding difference is less than or equal to the difference threshold, then the real-time operating data is determined to belong to the healthy cluster category set, meaning the air cooler is operating normally. Conversely, if the difference between the real-time operating data vector and the feature center points of all healthy cluster categories in the healthy cluster category set exceeds the preset difference threshold, then the real-time operating data is determined not to belong to any healthy cluster category in the healthy cluster category set, meaning the air cooler is operating abnormally, and subsequent alarm procedures are triggered.

[0042] S5. When the difference exceeds the set threshold, the air cooler is judged to be operating abnormally and an alarm is triggered.

[0043] Specifically, the system monitors the difference value in real time. If it exceeds a preset threshold, the air cooler is deemed to be malfunctioning, and an alarm signal is immediately triggered to the ship's monitoring center. The alarm signal simultaneously activates the audible and visual warning devices and records the abnormal event in the operation log database for traceability and analysis. The entire detection process is automatically executed by the anomaly detection module, requiring no manual intervention and ensuring real-time response to abnormal air cooler conditions.

[0044] Reference Figure 3 This is a schematic diagram of a system structure for detecting abnormalities in a ship's main engine air cooler, as described in an embodiment of this application.

[0045] The system for detecting anomalies in the air cooler of a ship's main engine includes a data acquisition module 1, a data preprocessing module 2, a training module 3, and an anomaly detection module 4.

[0046] The data acquisition module 1 contains a multi-type sensor network 11, including temperature, pressure and flow sensors, deployed on the ship's main engine. It continuously collects historical and real-time operating data and transmits them to the data preprocessing module 2 via the ship's data bus.

[0047] The data preprocessing module 2 is used to preprocess historical and real-time running data; the integrated computing unit 21 is connected to the data acquisition module 1 through the data bus, receives the raw historical / real-time running data from the data acquisition module 1, performs data cleaning and normalization operations, and outputs standardized historical / real-time running data to the training module 3 and the anomaly detection module 4.

[0048] Training module 3 is configured with a server-grade processor computing device 31 and an in-memory database 32. It connects to the data preprocessing module 2 via a data bus to obtain historical running data, performs tasks such as determining the number of input layer nodes, building the network, and training, generates healthy cluster categories, forms a set of healthy cluster categories representing the health status, and stores them in the in-memory database 32.

[0049] The anomaly detection module 4 connects the data preprocessing module 2 and the training module 3. It is used to input the preprocessed real-time running data into the trained adaptive resonance theory network, match the output real-time running data to the corresponding healthy cluster category, calculate the difference degree, and determine the air cooler is operating abnormally and trigger an alarm signal when the difference degree exceeds the preset threshold.

[0050] Reference Figure 4-5 The following is a description of another embodiment of the method and system provided in this implementation.

[0051] Based on the method and system for detecting abnormal pressure of ship main engines described in Embodiment 1, this Embodiment 2 adds some specific implementation methods.

[0052] In this embodiment, step S206 includes dividing the data into multiple consecutive intervals of difference values ​​to establish a multi-level response mechanism, calculating the difference between the health cluster category corresponding to the real-time running data and the set of health cluster categories, and determining the interval of difference values ​​in which the data belongs and the corresponding response level.

[0053] Step S207 includes triggering a warning signal and activating the ship monitoring system data recording function when the difference value is in the first interval; triggering an emergency alarm signal and activating automatic main engine load adjustment when the difference value is in the second interval; and triggering a shutdown protection signal and cutting off the main engine air supply circuit when the difference value is in the third interval; wherein the boundary values ​​of the first interval, the second interval, and the third interval are dynamically adjusted.

[0054] The process involves periodically calling a set of differences, composed of varying degrees, as the analysis sample. The distribution density function of this set is calculated, and a continuous probability distribution curve is generated using a kernel density estimation algorithm. The peak position and its offset direction of this curve are then identified. If the peak shifts towards increasing differences, it indicates an overall deterioration in the air cooler's operating condition. In this case, the boundary values ​​of the first, second, and third intervals are simultaneously increased to delay response triggering and avoid frequent false alarms. If the peak shifts towards decreasing differences, it indicates a stabilizing operating condition. In this case, the boundary values ​​of each interval are decreased to improve detection sensitivity. The adjustment range of the boundary values ​​is positively correlated with the peak shift range, ensuring that the multi-level response mechanism always adapts to the current operating conditions.

[0055] Specifically, the first interval is set as a low-risk abnormality interval, triggering an early warning signal to activate the ship's monitoring system's full-process historical data recording function, and simultaneously displaying an early warning prompt on the bridge; the second interval is set as a medium-risk abnormality interval, triggering an emergency alarm signal to trigger the main engine control system to initiate automatic load adjustment, smoothly reducing the main engine output power through a closed-loop feedback algorithm; the third interval is set as a high-risk abnormality interval, triggering a shutdown protection signal to drive the pneumatic actuator to cut off the main engine's air supply circuit, forcing a shutdown to avoid equipment damage. The boundary value dynamic adjustment is achieved by the adjustment module periodically calling the difference set, generating a probability distribution curve based on its statistical distribution characteristics, and automatically adjusting the interval boundaries according to the direction of the curve peak offset. If the peak value shifts towards a higher difference direction, the boundary values ​​of the three intervals are simultaneously moved upwards to suppress false alarms caused by operating condition fluctuations; if the peak value shifts towards a lower difference direction, the boundary values ​​are moved downwards to improve detection sensitivity. The boundary value adjustment range adaptively matches the peak offset degree, ensuring that the multi-level response mechanism always adapts to the current ship operating status.

[0056] Reference Figure 5 A system for detecting abnormalities in a ship's main engine air cooler also includes a response control module 5 and an adjustment module 6.

[0057] The response control module 5 is connected to the anomaly detection module 4 and is used to establish a multi-level response mechanism based on the range of difference values.

[0058] The adjustment module 6 is connected to the response control module 5 and is used to periodically call the difference set, calculate the distribution density function of the difference set, and dynamically adjust the boundary values ​​of the first interval, the second interval and the third interval according to the peak offset direction and amplitude of the distribution density function.

[0059] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0060] Unless otherwise defined, the technical or scientific terms used in this application shall have the ordinary meaning understood by one of ordinary skill in the art to which this application pertains. The terms "first," "second," "third," and similar terms used in this application specification and claims do not indicate any order, quantity, or importance, but are merely used to distinguish different components. The terms "an" or "a" and similar terms do not indicate a quantity limitation, but rather indicate the presence of at least one. The terms "comprising" or "including" and similar terms mean that the elements or objects preceding "comprising" or "including" encompass the elements or objects listed following "comprising" or "including" and their equivalents, and do not exclude other elements or objects. "Above," "below," "left," "right," etc., are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0061] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A method for detecting abnormalities in a marine main engine air cooler, characterized in that: include S1. Obtain historical operating data of the air cooler when it is in a healthy state; S2. Preprocess the historical operation data to obtain preprocessed historical operation data; S3. Construct an adaptive resonance theory network based on preprocessed historical running data, and repeat training until a set of health cluster categories representing health status is obtained. S4. After performing the same preprocessing on the real-time collected running data, input it into the trained adaptive resonance theory network to determine whether the real-time running data belongs to any one of the healthy cluster categories in the healthy cluster category set. S5. When it is determined that the real-time operating data does not belong to any health cluster category, the air cooler is deemed to be operating abnormally and an alarm is triggered.

2. The method for detecting abnormalities in a marine main engine air cooler according to claim 1, characterized in that: The preprocessing of historical operating data includes data cleaning, removing outliers and noisy data, and then normalizing the data.

3. The method for detecting abnormalities in a marine main engine air cooler according to claim 1, characterized in that: S3 includes constructing an adaptive resonance theory network; inputting each preprocessed historical running data into the adaptive resonance theory network, calculating the similarity between the input historical running data and existing healthy cluster categories; if the similarity with any of the healthy cluster categories reaches the warning parameter, then the historical running data is assigned to that healthy cluster category and the category weight is updated; if the similarity with all healthy cluster categories does not reach the warning parameter, then a new healthy cluster category is created and the current historical running data sample features are stored; iterating through all historical running data until the adaptive resonance theory network converges, forming a set of healthy cluster categories representing the health status.

4. The method for detecting abnormalities in a marine main engine air cooler according to claim 3, characterized in that: The warning parameters are dynamically adjusted based on the navigation environment parameters, specifically including real-time acquisition of wind speed and wave height of the ship's environment; dynamic calculation of warning parameters based on changes in wind speed and wave height; matching the corresponding learning rate based on the dynamic numerical range of the warning parameters; and inputting the warning parameters and learning rate into the adaptive resonance theory network to update the network's healthy clustering category generation and weight adjustment logic.

5. The method for detecting abnormalities in a marine main engine air cooler according to claim 4, characterized in that: The learning rate for dynamic numerical interval matching based on warning parameters includes defining the upper and lower limits of the critical interval using a piecewise function strategy; when the warning parameter is less than the lower limit of the critical interval, the learning rate is set to a constant high value; when the warning parameter is greater than the upper limit of the critical interval, the learning rate is set to a constant low value; when the dynamic warning parameter is between the lower and upper limits of the critical interval, the learning rate decreases linearly as the warning parameter increases.

6. The method for detecting abnormalities in a marine main engine air cooler according to claim 3, characterized in that: Determining whether real-time running data belongs to any health cluster category involves calculating the difference between the real-time running data and each health cluster category in the health cluster category set. The difference is the Euclidean distance between the real-time running data and the feature centers of each health cluster category in the health cluster category set. When the difference between the real-time running data and all health cluster categories in the health cluster category set exceeds a set threshold, the real-time running data is determined not to belong to any health cluster category. When the difference between the real-time running data and any health cluster in the health cluster category set is less than or equal to the set threshold, the real-time running data is determined to belong to that health cluster category.

7. The method for detecting abnormalities in a marine main engine air cooler according to claim 6, characterized in that: It also includes step S6: establishing a multi-level response mechanism based on the difference value range; when the difference value is in the first range, triggering an early warning signal and starting the ship monitoring system data recording function; when the difference value is in the second range, triggering an emergency alarm signal and activating automatic adjustment of the main engine load; when the difference value is in the third range, triggering a shutdown protection signal and cutting off the main engine air supply circuit; wherein the boundary values ​​of the first range, the second range and the third range are dynamically adjusted.

8. The method for detecting abnormalities in a marine main engine air cooler according to claim 7, characterized in that: The dynamic adjustment of the boundary values ​​of the first, second, and third intervals includes periodically calling the set of differences consisting of differences; calculating the distribution density function of the set of differences; and adjusting the boundary values ​​of each interval according to the peak offset direction and magnitude of the distribution density function.

9. A system for detecting abnormalities in a marine main engine air cooler, characterized in that: It includes a data acquisition module, a data preprocessing module, a training module, and an anomaly detection module; The data acquisition module is used to collect historical and real-time operating data of the ship's main engine air cooler. The data preprocessing module, connected to the data acquisition module, is used to preprocess historical and real-time operational data. The training module, connected to the data preprocessing module, is used to determine the number of input layer nodes of the adaptive resonance theory network, construct the adaptive resonance theory network structure, and train the network based on the preprocessed historical running data to form a set of healthy cluster categories representing the health status. The anomaly detection module connects the data preprocessing module and the training module. It is used to input the preprocessed real-time running data into the trained adaptive resonance theory network and determine whether the real-time running data belongs to any healthy cluster category in the healthy cluster category set.

Images