A fault diagnosis method and system applied to impressed current cathodic protection system

Abstract

The application discloses a kind of applied to the fault diagnosis method and system of impressed current cathodic protection system, comprising;Through the different types of sensor of pre-setting, real-time or periodically from ICCP monitoring system under different fault condition category acquisition multi-source time series data, i.e.sensory data sequence data;Data preprocessing is carried out to multi-source time series data to obtain pre-processing time series data;Fusion algorithm model based on long short-term memory network LSTM+Attention of deep learning classification model is constructed, and artificial intelligence diagnosis model is obtained by pre-processing time series data;The hybrid fault diagnosis model based on artificial intelligence diagnosis model and expert rule engine is constructed;According to hybrid fault diagnosis model, realize the fault diagnosis applied to impressed current cathodic protection system.The present application solves the technical problems of insufficient fault diagnosis accuracy and low efficiency in the fault diagnosis process of ICCP system in the prior art.

Description

Technical Field

[0001] This invention relates to the field of corrosion protection technology for marine engineering equipment, and in particular to a fault diagnosis method and system for impressed current cathodic protection systems. Background Technology

[0002] ICCP systems are a key technology for protecting large marine metal structures such as ships, offshore platforms, and offshore wind turbine foundations from electrochemical corrosion. However, current fault diagnosis technologies for ICCP systems have the following problems:

[0003] 1. Inadequate diagnostic mechanisms and low reliability: Current mainstream monitoring systems primarily rely on fixed threshold alarm logic. An alarm is triggered when the measured value of a sensor (such as a reference electrode) or the output parameter of the anode exceeds a preset range. This method cannot effectively distinguish between instantaneous data fluctuations caused by dynamic changes in the complex marine environment or sensor noise and actual equipment failures, resulting in a high false alarm rate. This increases the workload for maintenance personnel and reduces the overall reliability of the system. This is because traditional methods lack the ability to deeply analyze the temporal characteristics and multivariate correlations of data.

[0004] 2. Vague fault location and lack of root cause analysis: Traditional systems, when alarming, typically only indicate abnormal parameters but cannot deeply analyze and locate the root cause of the fault. For example, the system struggles to automatically distinguish whether the reading drift is caused by aging of the reference electrode or by actual underprotection due to localized coating damage to the structure. For complex faults caused by the coupling of multiple components, existing technologies lack effective analysis and location capabilities. The technical reason lies in the fact that its diagnostic logic is isolated and based on single-point parameters, rather than systematic.

[0005] 3. Passive Operation and Maintenance Mode with Delayed Response: Current operation and maintenance largely follows a passive "repair after failure" model, meaning that maintenance personnel only intervene after a failure has occurred and protection effectiveness may have already decreased. This model lacks the ability to predict equipment health degradation trends and cannot support condition-based maintenance (CBM) or predictive maintenance, potentially leaving structures at risk of corrosion for extended periods. This stems from the system's lack of ability to learn from historical data to predict equipment degradation patterns and make forward-looking predictions.

[0006] 4. Idle Data Value: ICCP systems continuously generate massive amounts of time-series data (such as current, voltage, and potential) and event logs during operation. This data contains valuable information about equipment operating status, degradation patterns, and fault characteristics, but it is currently underutilized. Operational and maintenance decisions largely rely on engineers' personal experience, lacking a data-driven, standardized decision support system.

[0007] Therefore, in order to solve the problems of insufficient accuracy and low efficiency in fault diagnosis of ICCP systems caused by the above technical issues, it is urgent to develop a fault diagnosis method and system for impressed current cathodic protection systems. Summary of the Invention

[0008] This invention provides a fault diagnosis method and system for impressed current cathodic protection systems to overcome the above-mentioned technical problems.

[0009] To achieve the above objectives, the technical solution of the present invention is as follows:

[0010] A fault diagnosis method for an impressed current cathodic protection system includes the following steps:

[0011] S1: Through preset different types of sensors, multi-source time-series data, i.e. sensor data sequence data, under different fault conditions are collected from the ICCP monitoring system in real time or periodically.

[0012] Furthermore, the multi-source timing data includes at least the output current / output voltage of each auxiliary anode, the measured potential of each reference electrode, and the power supply operating parameters;

[0013] S2: Perform data preprocessing on multi-source time series data to obtain preprocessed time series data;

[0014] S3: Construct a fusion algorithm model based on the deep learning classification model Long Short-Term Memory Network (LSTM) + Attention, and train the fusion algorithm model by preprocessing time series data to obtain an artificial intelligence diagnostic model;

[0015] S4: Construct a hybrid fault diagnosis model based on artificial intelligence diagnostic models and expert rule engines;

[0016] The hybrid fault diagnosis model includes an input module, an expert rule engine module, an artificial intelligence diagnosis module, and a comprehensive evaluation module.

[0017] The input module is used to input preprocessed time series data into the expert rule engine module and the artificial intelligence diagnostic model, respectively; the expert rule engine module is used to output a first diagnostic result based on expert experience values ​​and the preprocessed time series data; the artificial intelligence diagnostic module is used to output a second diagnostic result based on the preprocessed time series data.

[0018] The comprehensive evaluation module is used to obtain the comprehensive confidence level and the corresponding fault prediction result based on the first and second diagnostic results.

[0019] S5: Based on the hybrid fault diagnosis model, fault diagnosis is achieved for the impressed current cathodic protection system.

[0020] Furthermore, S2 specifically includes the following steps:

[0021] S21: Use the quantile scale inherent in the data to replace the variance for data standardization of multi-source time series data to obtain standardized time series data;

[0022] S22: Based on a set sliding window, extract sequence segments from standardized time-series data to obtain sample sequence data of a specified length;

[0023] S23: Perform KNN noise suppression processing on the sample sequence data to obtain denoised sample data;

[0024] S24: Perform KNN intelligent imputation on denoised sample data to obtain intelligent imputation data;

[0025] S25: Perform KNN outlier identification and repair on the intelligent completion data to obtain optimized sample data;

[0026] S26: Perform sample augmentation processing on the minority class samples in the optimized sample data to obtain preprocessed time series data.

[0027] Furthermore, the formula for obtaining the denoised sample data in S23 is as follows:

[0028] ,

[0029] ,

[0030] In the formula: Represents class sample sequence data With class sample sequence data The Euclidean distance between them; Indicates the number of sequence fragments in the sample class; Represents class sample sequence data The first in A sequence segment; Represents class sample sequence data The first in A sequence segment; This represents the denoised estimate; Represents class sample sequence data of A nearest neighbor candidate set; Indicates intermediate parameters; This indicates a design parameter to prevent division by zero.

[0031] The expression in S24 for intelligent KNN missing value completion of denoised sample data:

[0032] ,

[0033] In the formula: Indicates the timing of intelligent auto-completion. Missing data; This represents a pre-defined environmental feature vector; Represents the nearest neighbor set of the feature vectors of the pre-defined environment. The elements in.

[0034] Furthermore, S25 specifically includes the following steps:

[0035] S251: Obtain the local outlier factor based on the intelligent completion data, and its expression is as follows:

[0036] ,

[0037] ,

[0038] In the formula: This indicates sample sequence data in the intelligent completion data. Local outlier; Represents class sample sequence data Locally achievable density; Represents class sample sequence data Locally achievable density; Represents class sample sequence data Locally achievable density; Represents class sample sequence data With class sample sequence data The Euclidean distance between them; express Distance to The distance to one neighbor; Represents class sample sequence data of A nearest neighbor candidate set;

[0039] S252: Identify outliers in the intelligent completion data based on local outlier factors;

[0040] That is, to determine whether the local outlier factor corresponding to the sample sequence data meets the preset threshold. ;

[0041] If the local outlier factor is greater than a preset threshold If the local outlier factor is identified as an anomalous data class, the neighborhood weighted quantile of the sample sequence data class is used to replace the sample sequence data class, i.e., anomalous data repair.

[0042] If the local outlier factor is less than or equal to a preset threshold If the data is confirmed to be normal data, no further processing is required, and optimized sample data is obtained.

[0043] Furthermore, the expression for sample augmentation processing in S26 is as follows:

[0044] ,

[0045] In the formula: This represents the data after sample augmentation. Indicates the first in the optimized sample data Each class of sample sequence data; express Elements in; Represents class sample sequence data of A nearest neighbor candidate set; Indicate design parameters; It represents a 0-1 normal distribution.

[0046] Furthermore, the S3-based fusion algorithm model based on the deep learning classification model Long Short-Term Memory Network (LSTM) + Attention includes a dynamic feature extraction module, a Long Short-Term Memory Network (LSTM), an Attention module, a context vector feature fusion module, and a fault probability prediction module.

[0047] The dynamic feature extraction module is used to obtain the temporal feature vectors in the preprocessed time series data;

[0048] Furthermore, the time-series feature vector includes at least statistical domain features, time-domain features, and correlation features between sensor data sequence data;

[0049] Long Short-Term Memory (LSTM) networks are used to extract temporal dependency features from temporal feature vectors.

[0050] Furthermore, the formula for extracting time-dependent features is:

[0051]

[0052]

[0053]

[0054]

[0055] Where: Where: They represent the first Layer LSTM at time steps The activation values ​​of the input gate, forget gate, and output gate; Indicates the state of candidate cells; Indicates at time step Updated cell state; Indicates at time step cellular state; , , , Represents the weight matrix; , , Indicates the bias term; express Activation function; This represents element-wise multiplication. Represents the hyperbolic tangent activation function; Indicates time step Corresponding time dependency features; Indicates time step -1 corresponds to the time dependency feature; Represents time series feature vectors elements in ; Indicates the length of the time series;

[0056] The Attention module is used to calculate the attention weights of temporally dependent features by introducing an attention mechanism; its expression is as follows:

[0057] ,

[0058] ,

[0059] In the formula: Indicates time step Attention weights corresponding to time-dependent features; , , Indicates trainable parameters; Indicates transpose; Indicates time step Feature enhancement vectors corresponding to time-dependent features; Indicates time step Feature enhancement vectors corresponding to time-dependent features;

[0060] The context vector feature fusion module is used to fuse attention weights. Features related to time dependence Obtain the context vector Its expression is

[0061] ,

[0062] The failure probability prediction module is used to convert the context vector Input to fully connected layer and Classifier, to obtain predicted probabilities And the predicted probability The expression is

[0063] ,

[0064] In the formula: This represents the weight matrix of the fully connected layer; This represents the bias term of the fully connected layer.

[0065] Furthermore, the expression for the comprehensive evaluation module in S4 is:

[0066]

[0067] In the formula: Indicates the calculation of weights; This indicates the probability of expert rule judgment, i.e., the first diagnostic result; This indicates the probability predicted by the AI ​​diagnostic module, which is the second diagnostic result. This represents the decision probability output by the comprehensive evaluation module.

[0068] A fault diagnosis system for impressed current cathodic protection systems includes a data acquisition module, a preprocessing module, an artificial intelligence diagnostic model construction module, a hybrid fault diagnosis model construction module, and a model deployment module.

[0069] The data acquisition module is used to collect multi-source time-series data, i.e. sensor data sequence data, from the ICCP monitoring system in real time or periodically under different fault conditions through preset different types of sensors.

[0070] Furthermore, the multi-source timing data includes at least the output current / output voltage of each auxiliary anode, the measured potential of each reference electrode, and the power supply operating parameters;

[0071] The preprocessing module is used to preprocess multi-source time series data to obtain preprocessed time series data;

[0072] The AI ​​diagnostic model building module is used to construct a fusion algorithm model based on the deep learning classification model Long Short-Term Memory Network (LSTM) + Attention, and to train the fusion algorithm model by preprocessing time-series data to obtain an AI diagnostic model.

[0073] The hybrid fault diagnosis model construction module is used to build a hybrid fault diagnosis model based on artificial intelligence diagnosis model and expert rule engine. It includes an input module, an expert rule engine module, an artificial intelligence diagnosis module and a comprehensive evaluation module.

[0074] The input module is used to input preprocessed time series data into the expert rule engine module and the artificial intelligence diagnostic model, respectively; the expert rule engine module is used to output a first diagnostic result based on expert experience values ​​and the preprocessed time series data; the artificial intelligence diagnostic module is used to output a second diagnostic result based on the preprocessed time series data.

[0075] The comprehensive evaluation module is used to obtain the comprehensive confidence level and the corresponding fault prediction result based on the first and second diagnostic results.

[0076] The model deployment module is used to deploy hybrid fault diagnosis models to enable fault diagnosis in impressed current cathodic protection systems.

[0077] Beneficial effects: This invention provides a fault diagnosis method and system for impressed current cathodic protection systems. By constructing a hybrid fault diagnosis model based on an artificial intelligence diagnostic model and an expert rule engine, it solves the problem of high false alarm and missed alarm rates in traditional ICCP fault monitoring systems that rely on a single threshold alarm. The hybrid diagnostic engine of this invention can greatly improve the accuracy and efficiency of fault diagnosis for ICCP systems by combining a data-driven AI model with a knowledge-driven expert rule engine. Attached Figure Description

[0078] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0079] Figure 1 This is a flowchart of the fault diagnosis method of the present invention applied to an impressed current cathodic protection system;

[0080] Figure 2 This is a diagram illustrating the core technology roadmap for the fault diagnosis method of the impressed current cathodic protection system in this embodiment.

[0081] Figure 3 This is a schematic diagram of the intelligent operation and maintenance decision report interface in this embodiment;

[0082] Figure 4 This is a comparison of performance data between the method and system described in this embodiment and traditional solutions;

[0083] Figure 5 This is a simulation diagram of fault diagnosis in this embodiment. Detailed Implementation

[0084] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0085] ICCP (Internal Corrosion Protection) systems are a key technology for protecting large marine metal structures such as ships, offshore platforms, and offshore wind power foundations from electrochemical corrosion. This embodiment utilizes artificial intelligence, machine learning, and big data analytics to monitor and analyze the operational status of the ICCP system in real time. The aim is to achieve accurate fault diagnosis, root cause analysis, and intelligent operation and maintenance decision-making, thereby improving system reliability and maintenance efficiency.

[0086] This embodiment provides a fault diagnosis method for an impressed current cathodic protection system, such as... Figures 1 to 2 As shown, the specific steps include:

[0087] S1: Through preset different types of sensors, multi-source time-series data, i.e. sensor data sequence data, under different fault conditions are collected from the ICCP monitoring system in real time or periodically.

[0088] Furthermore, the multi-source timing data includes at least the output current / output voltage of each auxiliary anode, the measured potential of each reference electrode, and the power supply operating parameters;

[0089] Specifically, in this embodiment, three types of data are collected and integrated from the ICCP monitoring system in real time or periodically: operating parameter data: output current and output voltage of each auxiliary anode; measured potential of each reference electrode; and operating parameters of the power supply module.

[0090] Equipment file data: static information such as model, material, installation location, and design life of key components (e.g., anode, reference electrode);

[0091] Historical event data: The system's historical alarm records, operation logs, and maintenance records;

[0092] In this embodiment, the operating parameter data is used as multi-source time-series data, and the device file data and historical event data are used to generate historical statistical data of expert rule engine rules based on expert experience values.

[0093] The method and system described in this embodiment can be applied to all marine engineering equipment and facilities equipped with impressed current cathodic protection (ICCP) systems, including but not limited to:

[0094] Ships: All types of commercial ships, engineering vessels, and military vessels;

[0095] Offshore oil and gas platforms: jacket platforms, semi-submersible platforms, FPSOs, etc.;

[0096] Offshore wind power: fixed foundations (such as monopiles and jackets) and floating foundations;

[0097] Port and wharf facilities: steel pipe pile and steel sheet pile wharves;

[0098] Cross-sea bridge and undersea tunnel: underwater steel structure section;

[0099] Submarine pipeline;

[0100] S2: Perform data preprocessing on multi-source time series data to obtain preprocessed time series data; specifically including the following steps:

[0101] S21: Use the quantile scale inherent in the data to replace the variance for data standardization of multi-source time series data to obtain standardized time series data;

[0102] S22: Based on a set sliding window, extract sequence segments from standardized time-series data to obtain sample sequence data of a specified length;

[0103] S23: Perform KNN noise suppression processing on the sample sequence data to obtain denoised sample data;

[0104] Specifically, the formula for obtaining the denoised sample data is as follows:

[0105] ,

[0106] ,

[0107] In the formula: Represents class sample sequence data With class sample sequence data The Euclidean distance between them; Indicates the number of sequence fragments in the sample class; Represents class sample sequence data The first in A sequence segment; Represents class sample sequence data The first in A sequence segment; This represents the denoised estimate; Represents class sample sequence data of A nearest neighbor candidate set; Indicates intermediate parameters; Indicate design parameters to prevent division by zero;

[0108] S24: Perform KNN intelligent imputation on denoised sample data to obtain intelligent imputation data;

[0109] Specifically, the expression for intelligent KNN missing value completion of the denoised sample data is as follows:

[0110] ,

[0111] In the formula: Indicates the timing of intelligent auto-completion. Missing data; This represents a pre-defined environmental feature vector; Represents the nearest neighbor set of the feature vectors of the pre-defined environment. The elements in the feature space, and the environmental feature vector is obtained by sensor acquisition and processing; in this embodiment, when a short-term missing detection / packet loss occurs, the current segment is used as the element in the feature space. Nearest neighbor regression imputation is performed; for long window missing, a hierarchical KNN with "cross-channel + cross-time" is used (first cross-channel at the same location, then cross-time in the same channel) to reduce bias;

[0112] S25: Perform KNN outlier identification and repair on the intelligent completion data to obtain optimized sample data, specifically including the following steps:

[0113] S251: Obtain the local outlier factor based on the intelligent completion data, and its expression is as follows:

[0114] ,

[0115] ,

[0116] In the formula: This indicates sample sequence data in the intelligent completion data. Local outlier; Represents class sample sequence data Locally achievable density; Represents class sample sequence data Locally achievable density; Represents class sample sequence data Locally achievable density; Represents class sample sequence data With class sample sequence data The Euclidean distance between them; express Distance to The distance of a neighbor, Used to suppress the effects of being "too close" (reachable distance); Represents class sample sequence data of A nearest neighbor candidate set;

[0117] S252: Identify outliers in the intelligent completion data based on local outlier factors;

[0118] That is, to determine whether the local outlier factor corresponding to the sample sequence data meets the preset threshold. ;

[0119] If the local outlier factor is greater than a preset threshold If the local outlier factor is identified as an anomalous data class, the neighborhood weighted quantile of the sample sequence data class is used to replace the sample sequence data class, i.e., anomalous data repair.

[0120] If the local outlier factor is less than or equal to a preset threshold If so, the sample sequence data corresponding to the local outlier is confirmed as normal data and no processing is performed.

[0121] In this embodiment, the local density is used as the criterion to calculate the distance from the sample to its local density. The average distance of each nearest neighbor is compared with the expected distance in the neighborhood; those exceeding a threshold are marked as anomalies; suspected anomalies are reconstructed using "neighborhood consistency" (using...). (Weighted quantile replacement of nearest neighbors), and retain anomaly markers for log and root cause analysis;

[0122] S26: Perform sample augmentation processing on the minority class samples in the optimized sample data to obtain preprocessed time series data;

[0123] Specifically, the expression for sample augmentation processing in S26 is as follows:

[0124] ,

[0125] In the formula: This represents the data after sample augmentation. Indicates the first in the optimized sample data Each class of sample sequence data; express Elements in; Represents class sample sequence data of A nearest neighbor candidate set; Indicate design parameters; Represents a 0-1 normal distribution;

[0126] In this implementation, while ensuring physical rationality, KNN-SMOTE interpolation is used to perform a small amount of interpolation synthesis on rare but important operational sample fragments (limited to the original neighborhood convex hull) to improve the stability of downstream classification / prediction models in edge scenarios. All synthesized samples are marked with an "enhancement" tag and included in log management: a quality score (a weighted summary of noise level, missing data ratio, anomaly repair ratio, registration error, environmental correction magnitude, etc.) is output for each log record. The "raw-cleaned-enhanced" three-state data and processing metadata are written to the operation and maintenance security log for traceable auditing and model iteration. This embodiment ensures the quality of data input to the diagnostic engine by adding dedicated data preprocessing and quality enhancement steps before diagnosis. This allows for the stability and reliability of diagnostic results even in real industrial environments where sensor data quality fluctuates, enhancing engineering applicability.

[0127] In this embodiment, data processing employs a KNN-based interpolation algorithm to uniformly clean and enhance multi-source time-series data from multiple sensor sequences, including corrosion, electrochemical, electrical environment, and structural vibration data. This data serves as the standard input for subsequent model training and online prediction. Standardized time-series data is obtained through data standardization and windowing: robust standardization is performed (using the inherent quantile scale of the data instead of variance) to address the dimensional differences among the sensors. Then, a sample segment of length L is constructed using a pre-set sliding window to preserve timestamps and spatial location information for subsequent spatiotemporal registration. KNN noise suppression: for each timestamp / channel, a candidate set is constructed consisting of temporally adjacent (±r time steps) channels and similar channels from multiple sensor sources. Similarity is measured using Euclidean / Mahathano distance. The weighted median of the nearest neighbors is used as the denoising estimate, thereby achieving robust smoothing against impulse noise and polarization disturbances. This embodiment improves data quality by eliminating noise interference in the original data, which can ensure the accuracy and robustness of subsequent feature extraction and model diagnosis.

[0128] S3: Construct a fusion algorithm model based on the deep learning classification model Long Short-Term Memory Network (LSTM) + Attention, and train the fusion algorithm model by preprocessing time series data to obtain an artificial intelligence diagnostic model;

[0129] Specifically, the constructed fusion algorithm model based on the deep learning classification model Long Short-Term Memory Network (LSTM) + Attention includes a dynamic feature extraction module, a Long Short-Term Memory Network (LSTM), an Attention module, a context vector feature fusion module, and a fault probability prediction module.

[0130] The dynamic feature extraction module is used to obtain the temporal feature vectors in the preprocessed time series data;

[0131] Furthermore, the time-series feature vector includes at least statistical domain features, time-domain features, and correlation features between sensor data sequence data;

[0132] The statistical domain features include at least the mean, standard deviation, variance, kurtosis, and skewness of the data within the time window; the time domain features include at least the first difference (rate of change), trend slope, and fluctuation range of the data; the correlation features include at least the correlation coefficient or covariance between a specific sensor data sequence and other relevant sensor data sequences; in this embodiment, a sliding time window is used to extract a set of feature vectors from the preprocessed time series data, which is used to transform the original time series data into structured feature vectors that can effectively characterize the dynamic behavior and health status of the ICCP system, facilitating subsequent learning and classification by machine learning models;

[0133] Long Short-Term Memory (LSTM) networks are used to extract temporal dependency features from temporal feature vectors.

[0134] Furthermore, the formula for extracting time-dependent features is:

[0135]

[0136]

[0137]

[0138]

[0139] Where: Where: They represent the first Layer LSTM at time steps The activation values ​​of the input gate, forget gate, and output gate; Indicates the state of candidate cells; Indicates at time step Updated cell state; Indicates at time step cellular state; , , , Represents the weight matrix; , , Indicates the bias term; express Activation function; This represents element-wise multiplication. Represents the hyperbolic tangent activation function; Indicates time step Corresponding time dependency features; Indicates time step -1 corresponds to the time dependency feature; Represents time series feature vectors elements in ; Indicates the length of the time series; Indicates feature dimension;

[0140] To highlight key time-slice features, the Attention module introduces an attention mechanism to calculate the attention weights for time-dependent features, expressed as follows:

[0141] ,

[0142] ,

[0143] In the formula: Indicates time step Attention weights corresponding to time-dependent features; , , Indicates trainable parameters; Indicates transpose; Indicates time step Feature enhancement vectors corresponding to time-dependent features; Indicates time step Feature enhancement vectors corresponding to time-dependent features;

[0144] The context vector feature fusion module is used to fuse attention weights. Features related to time dependence Obtain the context vector , where the context vector By integrating the contributions of each time point in the data sequence, it can highlight the characteristics of anomalies in advance; and obtain the context vector. The expression is

[0145] ,

[0146] The failure probability prediction module is used to convert the context vector Input to fully connected layer and The classifier obtains the predicted probabilities of normal / abnormal. And the predicted probability The expression is

[0147] ,

[0148] In the formula: This represents the weight matrix of the fully connected layer; This represents the bias term of the fully connected layer;

[0149] In addition, an alert is triggered when the predicted probability of an anomaly category meets the following conditions:

[0150] ,

[0151] in: Indicates the warning threshold;

[0152] This embodiment describes a method for training a fusion algorithm model using preprocessed time-series data to obtain an artificial intelligence diagnostic model, specifically including:

[0153] S100: By dividing the preprocessed time series data into training and test sets according to a preset ratio;

[0154] S200: Train a pre-set neural network model using a training set to obtain a trained fusion algorithm model;

[0155] S300: Based on the model loss function, determine whether the model output of the trained fusion algorithm model has converged using the test set; if yes, use the trained fusion algorithm model as the artificial intelligence diagnostic model; if not, adaptively adjust the parameter weights of the trained fusion algorithm model based on backpropagation, and repeat step S200. The model loss function in this embodiment includes, but is not limited to, either the cross-entropy loss function or the mean squared error function.

[0156] This embodiment combines Temporal dependency modeling capabilities and The module's key features highlight its capabilities, enabling it to capture long-term dependencies in monitoring data and automatically identify critical moments before a fault occurs, thus improving the accuracy and interpretability of the early warning model. Compared to the simple LSTM method, it exhibits superior fault prediction performance under complex operating conditions.

[0157] S4: Construct a hybrid fault diagnosis model based on artificial intelligence diagnostic models and expert rule engines;

[0158] The hybrid fault diagnosis model includes an input module, an expert rule engine module, an artificial intelligence diagnosis module, and a comprehensive evaluation module.

[0159] The input module is used to input preprocessed time series data into the expert rule engine module and the artificial intelligence diagnostic model, respectively.

[0160] The expert rule engine module is used to output the first diagnostic result based on expert experience values ​​and preprocessed time-series data. Specifically, the expert rule engine solidifies expert knowledge and physical principles in the ICCP field into a structured "IF-THEN" logical rule base, and obtains historical statistical data of expert rule engine rules based on expert experience values ​​through recorded historical equipment archive data and historical event data. The expert rule engine module is used to set discrimination thresholds for faults with clear physical logic and clear judgment boundaries through historical statistical data to achieve rapid judgment of fault categories and simultaneously output the corresponding confidence level.

[0161] For example, the example rule of the expert rule engine module is: IF (the average output current of a certain anode approaches zero) AND (its average output voltage is greater than the preset high voltage threshold) THEN The diagnostic conclusion is: "The anode is open-circuit fault".

[0162] The artificial intelligence diagnostic module is used to output a second diagnostic result by preprocessing time-series data; the first and second diagnostic results include fault category and confidence level.

[0163] The comprehensive evaluation module is used to obtain the comprehensive confidence level and the corresponding fault prediction result based on the first and second diagnostic results.

[0164] Specifically, the expression for the comprehensive evaluation module is:

[0165]

[0166] In the formula: Indicates the calculation of weights; This indicates the probability of expert rule judgment, i.e., the first diagnostic result; This indicates the probability predicted by the AI ​​diagnostic module, which is the second diagnostic result. This represents the decision probability output by the comprehensive evaluation module; Indicates to The result is truncated to non-negativity and then normalized using the same operator.

[0167] In this embodiment, a comprehensive evaluation module is set up to comprehensively evaluate the fault category and confidence level output by the artificial intelligence model and the diagnostic conclusion triggered by the expert rule engine. This comprehensive evaluation module can output the final fault diagnosis conclusion and comprehensive confidence level based on weighted, voting or preset decision logic.

[0168] S5: Based on the hybrid fault diagnosis model, fault diagnosis is achieved for the impressed current cathodic protection system.

[0169] A fault diagnosis system for impressed current cathodic protection systems includes a data acquisition module, a preprocessing module, an artificial intelligence diagnostic model construction module, a hybrid fault diagnosis model construction module, and a model deployment module.

[0170] The data acquisition module is used to collect multi-source time-series data, i.e. sensor data sequence data, from the ICCP monitoring system in real time or periodically under different fault conditions through preset different types of sensors.

[0171] Furthermore, the multi-source timing data includes at least the output current / output voltage of each auxiliary anode, the measured potential of each reference electrode, and the power supply operating parameters;

[0172] The preprocessing module is used to preprocess multi-source time series data to obtain preprocessed time series data;

[0173] The AI ​​diagnostic model building module is used to construct a fusion algorithm model based on the deep learning classification model Long Short-Term Memory Network (LSTM) + Attention, and to train the fusion algorithm model by preprocessing time-series data to obtain an AI diagnostic model.

[0174] The hybrid fault diagnosis model construction module is used to build a hybrid fault diagnosis model based on artificial intelligence diagnosis model and expert rule engine. It includes an input module, an expert rule engine module, an artificial intelligence diagnosis module and a comprehensive evaluation module.

[0175] The input module is used to input preprocessed time series data into the expert rule engine module and the artificial intelligence diagnostic model, respectively; the expert rule engine module is used to output a first diagnostic result based on expert experience values ​​and the preprocessed time series data; the artificial intelligence diagnostic module is used to output a second diagnostic result based on the preprocessed time series data.

[0176] The comprehensive evaluation module is used to obtain the comprehensive confidence level and the corresponding fault prediction result based on the first and second diagnostic results.

[0177] The model deployment module is used to deploy hybrid fault diagnosis models to enable fault diagnosis in impressed current cathodic protection systems.

[0178] The specific implementation examples of the method and system described in this embodiment are as follows:

[0179] Case 1: Auxiliary Anode Open Circuit Fault Diagnosis. When the current of auxiliary anode 3 (A03) suddenly drops to 0A and the voltage rises to the upper limit, it is confirmed to be non-noise after preprocessing. The extracted features (mean current ≈ 0, mean voltage > threshold) are simultaneously input into two engines:

[0180] The expert rule engine module triggers the "open anode" rule, outputting the conclusion "A03 Open anode" with a confidence level of 95%.

[0181] AI models (such as LSTM + Attention fusion algorithm) identify this feature combination as "open anode" pattern and output the same conclusion with a probability of 90%.

[0182] The integrated judgment module confirms the consistent results and outputs the final diagnosis: "A03 is confirmed to have an open circuit fault", with a comprehensive confidence level of 93%.

[0183] Case 2: Reference Electrode Drift Fault Diagnosis. When the reading of reference electrode 2 (R02) was monitored to deviate slowly in one direction over several weeks, while the neighboring electrodes remained stable, its long-term trend slope and other characteristics were extracted after preprocessing.

[0184] AI models (such as LSTM + Attention fusion algorithms) identify this unique long-term, unidirectional, irreversible drift pattern and output "reference electrode drift" with a probability of 85%.

[0185] The expert rule engine module triggers the rule "monthly average drift rate > threshold and adjacent electrodes are stable", and outputs a similar conclusion with a confidence level of 80%.

[0186] The integrated judgment module, also known as the comprehensive evaluation module, makes a comprehensive judgment and outputs the final diagnosis: "R02 is confirmed to have a drift fault", with a comprehensive confidence level of 83%.

[0187] This embodiment also includes using existing technologies to transform diagnostic results into actionable guidelines, achieving an intelligent upgrade closed-loop operation and maintenance process from "alarm" to "decision support":

[0188] Workflow A: Once the diagnostic engine confirms a fault, the system automatically integrates information from multiple sources to generate structured operational and maintenance decision-making suggestions, such as... Figure 3 As shown. B. Information Integration: Integrates current diagnostic information, predictive information from related modules (such as the remaining useful life (RUL) of components), and standard operating procedures (SOPs) and spare parts information from the built-in expert knowledge base. C. Decision Generation: Automatically generates a maintenance work order or suggestion report that includes a fault description, root cause analysis, handling recommendations, and processing priorities.

[0189] Specific implementation examples:

[0190] Following Case 1, the system generates a decision report: Priority: Urgent. Root cause analysis: 1. Cable breakage; 2. Terminal corrosion. Recommended actions: 1. Immediately dispatch personnel to the site for investigation; 2. Temporarily increase the current of adjacent anodes by 5-10% to compensate for protection.

[0191] Following Case 2, the system integrates the predicted information of "R02 remaining lifetime < 90 days" and generates a decision report: Priority: Medium. Recommendations: 1. Plan to replace the electrode within 3 months; 2. The system automatically marks the R02 data as invalid and uses estimated values ​​for control.

[0192] This embodiment employs a hybrid fault diagnosis model: its core improvement lies in creatively combining data-driven AI models and knowledge-driven expert rules to overcome the limitations of single methods. AI models handle complex, non-linear problems, while expert rules handle explicit, physics-based problems. The two complement each other, significantly improving the accuracy and coverage of the diagnosis. Closed-loop intelligent decision-making: This embodiment goes beyond diagnosis, extending to automatically generating operational decisions that include root cause analysis, handling steps, and priorities, forming an intelligent closed loop of "diagnosis-decision-making," a fundamental transformation of the traditional passive alarm model. In addition, alternatives to the AI ​​diagnostic model in this embodiment include SVM models, random forest algorithms, etc., which can be interchanged with the LSTM + Attention fusion algorithm or combined using ensemble learning. Other alternatives, such as Gradient Boosting Decision Tree (GBDT) and Convolutional Neural Networks (CNN), can also be used depending on the specific fault type and data characteristics. The expert rule engine module in this embodiment is implemented based on Business Rule Management System (BRMS), but can also be custom logic judgment code. The method for constructing expert rules in the expert rule engine module is a well-known existing technology and will not be elaborated further here. Figure 4 The illustration demonstrates that the method and system described in this embodiment far surpass traditional methods in terms of diagnostic accuracy, timeliness, and low false alarm rate, enabling rapid and accurate fault location. Figure 5 The figure illustrates that the AI ​​model in the method and system described in this embodiment can predict future potential change trends based on historical data. The figure shows that the model predicts an abnormal potential drop in the T+1th cycle, thereby achieving early warning of faults such as "anode grounding" and providing decision support for preventative maintenance.

[0193] The beneficial effects of the method and system described in this embodiment are as follows: By constructing a hybrid fault diagnosis model based on an artificial intelligence diagnostic model and an expert rule engine, the problem of high false alarm and missed alarm rates caused by the reliance on a single threshold alarm in traditional ICCP fault monitoring systems is solved. The hybrid diagnostic engine of this invention can greatly improve the accuracy and efficiency of fault diagnosis of ICCP systems by combining a data-driven AI model with a knowledge-driven expert rule engine.

[0194] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A fault diagnosis method applied to an impressed current cathodic protection system, characterized in that, Specifically, the following steps are included: S1: Through preset different types of sensors, multi-source time-series data, i.e. sensor data sequence data, under different fault conditions are collected from the ICCP monitoring system in real time or periodically. Furthermore, the multi-source timing data includes at least the output current / output voltage of each auxiliary anode, the measured potential of each reference electrode, and the power supply operating parameters; S2: Perform data preprocessing on multi-source time series data to obtain preprocessed time series data; S3: Construct a fusion algorithm model based on the deep learning classification model Long Short-Term Memory Network (LSTM) + Attention, and train the fusion algorithm model by preprocessing time series data to obtain an artificial intelligence diagnostic model; S4: Construct a hybrid fault diagnosis model based on artificial intelligence diagnostic models and expert rule engines; The hybrid fault diagnosis model includes an input module, an expert rule engine module, an artificial intelligence diagnosis module, and a comprehensive evaluation module. The input module is used to input preprocessed time series data into the expert rule engine module and the artificial intelligence diagnostic model, respectively; the expert rule engine module is used to output the first diagnostic result based on expert experience values ​​and the preprocessed time series data. The artificial intelligence diagnostic module is used to output a second diagnostic result by preprocessing time-series data; The comprehensive evaluation module is used to obtain the comprehensive confidence level and the corresponding fault prediction result based on the first and second diagnostic results. S5: Based on the hybrid fault diagnosis model, fault diagnosis is achieved for the impressed current cathodic protection system.

2. The fault diagnosis method for an impressed current cathodic protection system according to claim 1, characterized in that, S2 specifically includes the following steps: S21: Use the quantile scale inherent in the data to replace the variance for data standardization of multi-source time series data to obtain standardized time series data; S22: Based on a set sliding window, extract sequence segments from standardized time-series data to obtain sample sequence data of a specified length; S23: Perform KNN noise suppression processing on the sample sequence data to obtain denoised sample data; S24: Perform KNN intelligent imputation on denoised sample data to obtain intelligent imputation data; S25: Perform KNN outlier identification and repair on the intelligent completion data to obtain optimized sample data; S26: Perform sample augmentation processing on the minority class samples in the optimized sample data to obtain preprocessed time series data.

3. The fault diagnosis method for an impressed current cathodic protection system according to claim 2, characterized in that, The formula for obtaining the denoised sample data in S23 is: In the formula: Represents class sample sequence data With class sample sequence data The Euclidean distance between them; Indicates the number of sequence fragments in the sample class; Represents class sample sequence data The first in A sequence segment; Represents class sample sequence data The first in A sequence segment; This represents the denoised estimate; Represents class sample sequence data of A nearest neighbor candidate set; Indicates intermediate parameters; This indicates a design parameter to prevent division by zero. The expression for intelligent KNN missing value completion of denoised sample data described in S24 is: In the formula: Indicates the timing of intelligent auto-completion. Missing data; This represents a pre-defined environmental feature vector; Represents the nearest neighbor set of the feature vectors of the pre-defined environment. The elements in.

4. The fault diagnosis method for an impressed current cathodic protection system according to claim 3, characterized in that, S25 specifically includes the following steps: S251: Obtain the local outlier factor based on the intelligent completion data, and its expression is as follows: In the formula: This indicates sample sequence data in the intelligent completion data. Local outlier; Represents class sample sequence data Locally achievable density; Represents class sample sequence data Locally achievable density; Represents class sample sequence data Locally achievable density; Represents class sample sequence data With class sample sequence data The Euclidean distance between them; express Distance to The distance to one neighbor; Represents class sample sequence data of A nearest neighbor candidate set; S252: Identify outliers in the intelligent completion data based on local outlier factors; That is, to determine whether the local outlier factor corresponding to the sample sequence data meets the preset threshold. ; If the local outlier factor is greater than a preset threshold If the local outlier factor is identified as an anomalous data class, the neighborhood weighted quantile of the sample sequence data class is used to replace the sample sequence data class, i.e., anomalous data repair. If the local outlier factor is less than or equal to a preset threshold If the data is confirmed to be normal data, no further processing is required, and optimized sample data is obtained.

5. A fault diagnosis method for an impressed current cathodic protection system according to claim 4, characterized in that, The expression for sample augmentation in S26 is: In the formula: This represents the data after sample augmentation. Indicates the first in the optimized sample data Each class of sample sequence data; express Elements in; Represents class sample sequence data of A nearest neighbor candidate set; Indicate design parameters; It represents a 0-1 normal distribution.

6. The fault diagnosis method for an impressed current cathodic protection system according to claim 4, characterized in that, The S3 model is a fusion algorithm based on the deep learning classification model Long Short-Term Memory Network (LSTM) + Attention, which includes a dynamic feature extraction module, a Long Short-Term Memory Network (LSTM), an Attention module, a context vector feature fusion module, and a fault probability prediction module. The dynamic feature extraction module is used to obtain the temporal feature vectors in the preprocessed time series data; Furthermore, the time-series feature vector includes at least statistical domain features, time-domain features, and correlation features between sensor data sequence data; Long Short-Term Memory (LSTM) networks are used to extract temporal dependency features from temporal feature vectors. Furthermore, the formula for extracting time-dependent features is: Where: Where: They represent the first Layer LSTM at time steps The activation values ​​of the input gate, forget gate, and output gate; Indicates the state of candidate cells; Indicates at time step Updated cell state; Indicates at time step cellular state; , , , Represents the weight matrix; , , Indicates the bias term; express Activation function; This represents element-wise multiplication. Represents the hyperbolic tangent activation function; Indicates time step Corresponding time dependency features; Indicates time step -1 corresponds to the time dependency feature; Represents time series feature vectors elements in ; Indicates the length of the time series; The Attention module is used to calculate the attention weights of temporally dependent features by introducing an attention mechanism; its expression is as follows: In the formula: Indicates time step Attention weights corresponding to time-dependent features; , , Indicates trainable parameters; Indicates transpose; Indicates time step Feature enhancement vectors corresponding to time-dependent features; Indicates time step Feature enhancement vectors corresponding to time-dependent features; The context vector feature fusion module is used to fuse attention weights. Features related to time dependence Obtain the context vector Its expression is The failure probability prediction module is used to convert the context vector Input to fully connected layer and Classifier, to obtain predicted probabilities And the predicted probability The expression is In the formula: This represents the weight matrix of the fully connected layer; This represents the bias term of the fully connected layer.

7. A fault diagnosis method for an impressed current cathodic protection system according to claim 6, characterized in that, The expression for the comprehensive evaluation module in S4 is: In the formula: Indicates the calculation of weights; This indicates the probability of expert rule judgment, i.e., the first diagnostic result; This indicates the probability predicted by the AI ​​diagnostic module, which is the second diagnostic result. This represents the decision probability output by the comprehensive evaluation module.

8. A system based on the fault diagnosis method for an impressed current cathodic protection system according to any one of claims 1 to 7, characterized in that, It includes a data acquisition module, a preprocessing module, an artificial intelligence diagnostic model building module, a hybrid fault diagnosis model building module, and a model deployment module; The data acquisition module is used to collect multi-source time-series data, i.e. sensor data sequence data, from the ICCP monitoring system in real time or periodically under different fault conditions through preset different types of sensors. Furthermore, the multi-source timing data includes at least the output current / output voltage of each auxiliary anode, the measured potential of each reference electrode, and the power supply operating parameters; The preprocessing module is used to preprocess multi-source time series data to obtain preprocessed time series data; The AI ​​diagnostic model building module is used to construct a fusion algorithm model based on the deep learning classification model Long Short-Term Memory Network (LSTM) + Attention, and to train the fusion algorithm model by preprocessing time-series data to obtain an AI diagnostic model. The hybrid fault diagnosis model construction module is used to build a hybrid fault diagnosis model based on artificial intelligence diagnosis model and expert rule engine. It includes an input module, an expert rule engine module, an artificial intelligence diagnosis module and a comprehensive evaluation module. The input module is used to input preprocessed time series data into the expert rule engine module and the artificial intelligence diagnostic model, respectively; the expert rule engine module is used to output the first diagnostic result based on expert experience values ​​and the preprocessed time series data. The artificial intelligence diagnostic module is used to output a second diagnostic result by preprocessing time-series data; The comprehensive evaluation module is used to obtain the comprehensive confidence level and the corresponding fault prediction result based on the first and second diagnostic results. The model deployment module is used to deploy hybrid fault diagnosis models to enable fault diagnosis in impressed current cathodic protection systems.

Images