An artificial intelligence-based marine equipment life cycle visual monitoring method

Abstract

The application provides a marine equipment life cycle visual monitoring method based on artificial intelligence, and belongs to the technical field of marine equipment.The application realizes intelligent filtering of environmental interference by collecting multi-source monitoring data, using a marine circulation cycle feedback filtering model to adaptively adjust potential space sampling randomness scale parameters according to the marine circulation cycle interference intensity, dynamically determining filtering model working parameters through a marine circulation cycle adjustment function to ensure the best filtering effect under different marine environmental conditions, extracting pure equipment state characteristic parameters such as vibration acceleration amplitude change rate, temperature gradient change rate, stress concentration coefficient and corrosion rate, and using a time sequence deep learning algorithm to analyze state characteristic evolution rules and predict equipment degradation trends, so that the technical problem of low monitoring precision caused by environmental interference in the marine equipment life cycle state monitoring process is solved.

Description

Technical Field

[0001] This invention belongs to the field of marine equipment technology, and more specifically, relates to an artificial intelligence-based method for visual monitoring of the life cycle of marine equipment. Background Technology

[0002] Lifecycle monitoring technology for marine equipment is crucial for ensuring the safe operation and extending the service life of marine engineering equipment. Traditional monitoring methods primarily employ single-sensor monitoring based on threshold judgment, periodic inspections, and condition assessment techniques based on empirical models. In complex engineering applications such as offshore oil drilling platforms, offshore wind power generation equipment, subsea pipeline systems, and deep-sea operation equipment, existing technologies collect data by deploying vibration sensors, pressure sensors, and temperature sensors, and use traditional frequency domain analysis and time domain statistical methods for signal processing and condition identification. However, traditional monitoring technologies face severe environmental interference problems in complex marine environments. Multiple environmental factors, such as ocean currents, wave impacts, seawater temperature fluctuations, and seabed sediment disturbances, introduce a large amount of noise and spurious signals into the monitoring signals. Traditional fixed-parameter filtering methods, such as bandpass filtering and low-pass filtering, cannot effectively separate environmental interference from the true equipment condition signal. Due to the strong time-varying and complex nature of the marine environment, traditional filtering techniques cannot adaptively adjust filtering parameters according to changes in environmental conditions. The filtering effect decreases significantly under strong currents or severe sea conditions, resulting in environmental interference dominating the monitoring data, severely masking the true equipment condition information, and drastically reducing monitoring accuracy. Summary of the Invention

[0003] In view of this, the present invention provides an artificial intelligence-based visualization monitoring method for the life cycle of marine equipment, which can solve the technical problem of low monitoring accuracy caused by environmental interference in the life cycle status monitoring of marine equipment in the prior art.

[0004] This invention is implemented as follows: It provides an artificial intelligence-based method for visual monitoring of the lifecycle of marine equipment, comprising: deploying a multi-dimensional sensor array on the marine equipment; collecting multi-source monitoring data during the operation of the marine equipment and performing time synchronization processing and data preprocessing to obtain synchronous multi-source monitoring data; using an ocean circulation feedback filtering model to filter out environmental interference from the synchronous multi-source monitoring data; extracting marine equipment state characteristic parameters, establishing an equipment state characteristic vector matrix and performing normalization processing; establishing a marine equipment degradation trend prediction model, using a time-series deep learning algorithm to analyze the evolution law of the equipment state characteristic vector matrix, predicting the performance degradation trend and failure probability of the equipment in future time periods, and outputting the equipment degradation prediction results; calculating the equipment lifecycle assessment index based on the equipment degradation prediction results and executing corresponding monitoring strategies according to the index range; constructing a three-dimensional visualization model of the marine equipment lifecycle, displaying the synchronous multi-source monitoring data, equipment state characteristic vector matrix, equipment degradation prediction results, and equipment lifecycle assessment index in real time using color mapping and dynamic charts, forming an intuitive display interface of the equipment's full lifecycle status, and dynamically adjusting the monitoring system parameters through an ocean circulation feedback control algorithm.

[0005] The multi-dimensional sensor array step involves deploying vibration acceleration sensors, temperature sensors, stress and strain sensors, and seawater corrosion monitoring sensors at key parts of the marine equipment. Specifically, vibration acceleration sensors are deployed at the main bearing of the equipment, temperature sensors are deployed in the heat dissipation area of ​​the equipment, stress and strain sensors are deployed in the load-bearing structure of the equipment, and seawater corrosion monitoring sensors are deployed on the seawater contact surface of the equipment, forming a multi-parameter monitoring network covering the entire life cycle of the equipment.

[0006] The steps for collecting multi-source monitoring data are as follows: vibration acceleration signals are continuously collected from the triaxial acceleration data of the main bearing of the equipment using a vibration acceleration sensor at a sampling frequency of 1000Hz; temperature change curves are continuously monitored from the temperature data of the heat dissipation area of ​​the equipment using a temperature sensor at a sampling frequency of 1Hz; stress distribution data are measured in real time from the strain value of the load-bearing structure of the equipment using a stress-strain sensor at a sampling frequency of 100Hz and converted into stress data; and corrosion potential change data are monitored from the electrochemical corrosion potential of the seawater contact surface of the equipment using a seawater corrosion monitoring sensor at a sampling frequency of 0.1Hz.

[0007] The steps of the ocean circulation feedback filtering model for filtering environmental interference are as follows: the filtering parameters are automatically adjusted according to the ocean circulation intensity. When the ocean circulation interference coefficient ∈ (0.6, 1.0], wavelet denoising is added; when the ocean circulation interference coefficient ∈ [0.3, 0.6], standard filtering is used; and when the ocean circulation interference coefficient ∈ [0, 0.3], mild filtering is used.

[0008] Specifically, the steps for extracting the state characteristic parameters of marine equipment are as follows: the rate of change of vibration acceleration amplitude is obtained by performing time-domain statistical analysis on the vibration acceleration signal and calculating the rate of change of the root mean square value of vibration acceleration within a continuous time window; the rate of change of temperature gradient is obtained by performing first-order difference operation on the temperature change curve, calculating the difference of temperature values ​​at adjacent time points and then dividing by the time interval; the stress concentration factor is obtained by performing spatial statistical analysis on the stress distribution data and calculating the ratio of the maximum stress value to the average stress value; and the corrosion rate is obtained by performing linear fitting analysis on the corrosion potential change data, calculating the slope of the corrosion potential change with time and converting it into the corrosion depth rate.

[0009] The steps for implementing the monitoring strategy based on the equipment life cycle assessment index are as follows: when the equipment life cycle assessment index is in the range of [0.8, 1.0), the enhanced monitoring mode is activated to increase the data collection frequency; when the equipment life cycle assessment index is in the range of [0.5, 0.8), a pre-maintenance alarm is triggered; when the equipment life cycle assessment index is in the range of [0.2, 0.5), the maintenance plan formulation procedure is initiated; and when the equipment life cycle assessment index is in the range of [0, 0.2), an equipment replacement recommendation is issued.

[0010] Among them, the ocean circulation feedback filtering model is a deep filtering network based on the variational autoencoder architecture, which includes an encoder layer, a latent space processing layer and a decoder layer. The encoder maps multidimensional monitoring signals to a low-dimensional latent space. The latent space processing layer uses the sampling randomness scale parameter of the latent space to adaptively separate the signal features. The decoder reconstructs the pure equipment status signal after removing environmental interference.

[0011] Among them, the potential space sampling random scale parameter is dynamically determined by the ocean circulation cycle adjustment function based on three parameters: ocean circulation velocity, ocean temperature variation amplitude, and equipment vibration reference value. Specifically, the ocean circulation velocity is obtained by measuring the current velocity vector of the water flow at the current location through a current meter installed near the marine equipment and calculating its modulus. The ocean temperature variation amplitude is obtained by analyzing the temperature data of the temperature sensor over the past 24 hours and calculating the difference between the maximum and minimum temperatures. The equipment vibration reference value is obtained by statistically analyzing the vibration data of the vibration acceleration sensor during the first 100 hours of normal operation of the equipment and calculating the root mean square value.

[0012] Specifically, the ocean circulation regulation function is calculated by weighting three parameters—ocean circulation velocity, ocean temperature variation amplitude, and equipment vibration reference value—to obtain the ocean circulation disturbance intensity value. When the ocean circulation disturbance intensity value ∈ [0, 0.3), the potential space sampling randomness scaling coefficient is set to 0.1; when the ocean circulation disturbance intensity value ∈ [0.3, 0.6), the potential space sampling randomness scaling coefficient is set to 0.3; and when the ocean circulation disturbance intensity value ∈ [0.6, 1.0], the potential space sampling randomness scaling coefficient is set to 0.8.

[0013] The ocean circulation feedback control algorithm specifically includes a parameter input module, a feedback calculation module, and a parameter adjustment output module. The parameter input module receives the rate of change of vibration acceleration amplitude, the rate of change of temperature gradient, the stress concentration factor, and the corrosion rate as input parameters. The feedback calculation module calculates the monitoring system performance evaluation index based on the input parameters using a weighted summation algorithm. The parameter adjustment output module outputs data acquisition frequency adjustment commands, filter intensity adjustment commands, and early warning threshold adjustment commands based on the monitoring system performance evaluation index.

[0014] Before establishing the ocean circulation feedback filtering model, the process includes building a training dataset. Specifically, this involves collecting operational data of marine equipment under different ocean circulation conditions, including multi-sensor monitoring signals under three typical operating conditions: weak, moderate, and strong ocean circulation. Data for each condition is manually labeled to mark the boundary between environmental interference signals and the equipment's true state signals, resulting in a standard training sample set containing 10,000 combinations of different ocean circulation intensities and equipment states. The next step is model training. This involves end-to-end training of the variational autoencoder using supervised learning. The loss function combines reconstruction error and latent space regularization. Network parameters are optimized using backpropagation. An adaptive learning rate scheduling strategy is employed during training, with an initial learning rate of 0.001. The learning rate decays to 0.8 times its original value every 50 training cycles, for a total of 500 training cycles. A stochastic gradient descent optimizer with a batch size of 32 is used.

[0015] Specifically, the weighted summation algorithm uses a weight of 0.3 for the rate of change of vibration acceleration amplitude, 0.2 for the rate of change of temperature gradient, 0.3 for the stress concentration factor, and 0.2 for the corrosion rate. The threshold comparison algorithm calculates the parameter adjustment coefficient based on the difference between the monitoring system performance evaluation index and the preset performance threshold.

[0016] Among them, the multi-dimensional sensor array refers to a combination of sensors installed in different parts of marine equipment to monitor the operating status of the equipment, including acceleration sensors for detecting mechanical vibration, temperature sensors for monitoring changes in environmental and equipment temperature, strain sensors for measuring structural stress distribution, and electrochemical sensors for assessing the degree of seawater corrosion.

[0017] Among them, the equipment life cycle assessment index refers to a quantitative evaluation index that comprehensively considers the current operating status of the equipment, historical degradation trends and predicted performance degradation. The value ranges from 0 to 1, and the higher the value, the better the equipment condition.

[0018] Among them, the monitoring system performance evaluation index refers to a comprehensive indicator for evaluating the overall working effect and data quality of the monitoring system, with a value range of 0 to 1. The parameter adjustment demand coefficient refers to a quantitative indicator for determining the necessity and adjustment range of system parameters based on the current monitoring status.

[0019] This invention establishes an intelligent filtering mechanism capable of adaptively separating environmental interference from equipment status signals by constructing a marine circulation cyclic feedback filtering model based on a variational autoencoder. This effectively solves the problem of poor filtering performance of traditional fixed-parameter filtering methods in complex marine environments. The filtering model utilizes the nonlinear mapping capability of deep learning networks to map multidimensional monitoring signals to a low-dimensional latent space for feature separation. An adaptive sampling mechanism in the latent space processing layer dynamically adjusts the signal separation parameters according to the intensity of the marine circulation cycle, achieving accurate identification and effective filtering of environmental interference of varying intensities. The marine circulation cycle adjustment function dynamically determines the operating parameters of the filtering model based on real-time marine circulation velocity, temperature variation amplitude, and equipment vibration reference values. It automatically switches the filtering mode when the intensity of environmental interference changes, ensuring optimal signal purification under various marine environmental conditions and significantly improving the signal-to-noise ratio and the accuracy of status information extraction. In summary, this invention solves the technical problem mentioned in the background art of low monitoring accuracy caused by environmental interference during the life-cycle status monitoring of marine equipment. Attached Figure Description

[0020] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0021] 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.

[0022] like Figure 1 The diagram shown is a flowchart of a visualization monitoring method for the life cycle of marine equipment based on artificial intelligence, provided by this invention. This method includes the following steps:

[0023] S01. Deploy multi-dimensional sensor arrays in key parts of marine equipment, including vibration acceleration sensors, temperature sensors, stress and strain sensors and seawater corrosion monitoring sensors. Vibration acceleration sensors are deployed in the main bearing of the equipment, temperature sensors are deployed in the heat dissipation area of ​​the equipment, stress and strain sensors are deployed in the load-bearing structure of the equipment, and seawater corrosion monitoring sensors are deployed in the seawater contact surface of the equipment, forming a multi-parameter monitoring network covering the entire life cycle of the equipment.

[0024] S02. Collect multi-source monitoring data during the operation of marine equipment. Among them, vibration acceleration signal is continuously collected by vibration acceleration sensor at a sampling frequency of 1000Hz for triaxial acceleration data of the main bearing of the equipment. Temperature change curve is continuously monitored by temperature sensor at a sampling frequency of 1Hz for temperature data of the heat dissipation area of ​​the equipment. Stress distribution data is measured in real time by stress strain sensor at a sampling frequency of 100Hz for strain value of the load-bearing structure of the equipment and converted into stress data. Corrosion potential change data is monitored by seawater corrosion monitoring sensor at a sampling frequency of 0.1Hz for electrochemical corrosion potential of the seawater contact surface of the equipment. The collected raw monitoring data is processed by time synchronization and data preprocessing to obtain synchronous multi-source monitoring data.

[0025] S03. Use the ocean circulation feedback filtering model to filter out environmental interference from the synchronous multi-source monitoring data. Automatically adjust the filtering parameters according to the ocean circulation intensity. When the ocean circulation interference coefficient ∈ (0.6, 1.0], add wavelet denoising processing. When the ocean circulation interference coefficient ∈ [0.3, 0.6], use standard filtering processing. When the ocean circulation interference coefficient ∈ [0, 0.3), use mild filtering processing.

[0026] S04. Extract the state characteristic parameters of marine equipment. The vibration acceleration amplitude change rate is obtained by performing time-domain statistical analysis on the vibration acceleration signal and calculating the change rate of the root mean square value of vibration acceleration within a continuous time window. The temperature gradient change rate is obtained by performing first-order difference operation on the temperature change curve, calculating the difference of temperature values ​​at adjacent time points and then dividing by the time interval. The stress concentration factor is obtained by performing spatial statistical analysis on the stress distribution data and calculating the ratio of the maximum stress value to the average stress value. The corrosion rate is obtained by performing linear fitting analysis on the corrosion potential change data, calculating the slope of the corrosion potential change with time and converting it into the corrosion depth rate. Establish the equipment state characteristic vector matrix and perform normalization processing.

[0027] S05. Establish a marine equipment degradation trend prediction model, use time series deep learning algorithm to analyze the evolution law of the equipment state feature vector matrix, predict the performance degradation trend and failure probability of the equipment in the future time period, and output the equipment degradation prediction results.

[0028] S06. Calculate the equipment life cycle assessment index based on the equipment degradation prediction results. When the equipment life cycle assessment index is in the range of [0.8, 1.0), activate the enhanced monitoring mode to increase the data collection frequency. When the equipment life cycle assessment index is in the range of [0.5, 0.8), trigger the pre-maintenance alarm. When the equipment life cycle assessment index is in the range of [0.2, 0.5), activate the maintenance plan formulation program. When the equipment life cycle assessment index is in the range of [0, 0.2), issue an equipment replacement recommendation.

[0029] S07. Construct a three-dimensional visualization model of the marine equipment life cycle, and display the synchronous multi-source monitoring data, the equipment status feature vector matrix, the equipment degradation prediction results, and the equipment life cycle assessment index in real time in the form of color mapping and dynamic charts to form an intuitive display interface of the equipment's full life cycle status, and realize the dynamic adjustment of monitoring system parameters through the ocean circulation feedback control algorithm.

[0030] The specific structure of the ocean circulation cyclic feedback filtering model is a deep filtering network based on a variational autoencoder architecture, comprising an encoder layer, a latent space processing layer, and a decoder layer. The encoder maps multidimensional monitoring signals to a low-dimensional latent space. The latent space processing layer adaptively separates signal features using latent space sampling randomness scale parameters. The decoder reconstructs the pure equipment state signal after removing environmental interference. The latent space sampling randomness scale parameters are dynamically determined based on three parameters: ocean circulation velocity, ocean temperature variation amplitude, and equipment vibration reference value, through an ocean circulation cyclic adjustment function. The steps for establishing the training dataset for the ocean circulation cyclic feedback filtering model specifically include collecting marine equipment operation data under different ocean circulation conditions, including multi-sensor monitoring signals under three typical operating conditions: weak circulation sea state, moderate circulation sea state, and strong circulation sea state. Data under each operating condition is manually labeled to mark the boundary between environmental interference signals and the true equipment state signals. A standard training sample set containing 10,000 sets of different ocean circulation intensities and equipment state combinations is established, with each sample containing the original mixed signal and the corresponding pure state signal label. The training steps for the ocean circulation cyclic feedback filtering model specifically include: end-to-end training of the variational autoencoder using supervised learning; combining reconstruction error and latent space regularization terms in the loss function; optimizing network parameters through backpropagation; employing an adaptive learning rate scheduling strategy during training, with an initial learning rate of 0.001, which decays to 0.8 times its original value every 50 training epochs, for a total of 500 training epochs; using a stochastic gradient descent optimizer with a batch size of 32; and testing the model performance on a validation set after training to ensure a filtering accuracy of over 95%.

[0031] The ocean circulation cycle adjustment function is used to adjust the potential spatial sampling stochastic scale parameter of the ocean circulation cycle feedback filter model. The function is calculated based on three data points: ocean circulation velocity, ocean temperature variation amplitude, and equipment vibration reference value. The ocean circulation velocity is obtained by measuring the current velocity vector at the current location using a current meter installed near the marine equipment and calculating its modulus. The ocean temperature variation amplitude is obtained by analyzing temperature data from the temperature sensor over the past 24 hours and calculating the difference between the maximum and minimum temperatures. The equipment vibration reference value is calculated by statistically analyzing the vibration data from the vibration acceleration sensor during the first 100 hours of normal equipment operation and calculating the mean square. The root value is obtained by calculating an ocean circulation disturbance intensity value using the weighted average formula of the three parameters. When the ocean circulation disturbance intensity value falls within different ranges, different potential spatial sampling random scaling coefficients are used to adjust the potential spatial sampling parameters of the ocean circulation feedback filter model. When the ocean circulation disturbance intensity value ∈ [0, 0.3), the potential spatial sampling random scaling coefficient is set to 0.1; when the ocean circulation disturbance intensity value ∈ [0.3, 0.6), the potential spatial sampling random scaling coefficient is set to 0.3; and when the ocean circulation disturbance intensity value ∈ [0.6, 1.0], the potential spatial sampling random scaling coefficient is set to 0.8.

[0032] The ocean circulation feedback control algorithm automatically optimizes system performance by establishing a cyclic feedback relationship between monitoring parameters. The algorithm includes a parameter input module, a cyclic feedback calculation module, and a parameter adjustment output module. The parameter input module receives the vibration acceleration amplitude change rate, the temperature gradient change rate, the stress concentration factor, and the corrosion rate as input parameters. The cyclic feedback calculation module calculates the monitoring system performance evaluation index based on the input parameters using a weighted summation algorithm, where the vibration acceleration amplitude change rate has a weight of 0.3, the temperature gradient change rate has a weight of 0.2, and the stress concentration factor has a weight of 0.3. The corrosion rate weight is 0.2. Then, the parameter adjustment requirement coefficient is calculated based on the difference between the monitoring system performance evaluation index and the preset performance threshold using a threshold comparison algorithm. The parameter adjustment output module outputs data acquisition frequency adjustment instructions, filter intensity adjustment instructions, and early warning threshold adjustment instructions based on the monitoring system performance evaluation index and the parameter adjustment requirement coefficient using a lookup table mapping method. The data acquisition frequency adjustment instruction is used to adjust the data acquisition frequency of the multi-dimensional sensor array. The filter intensity adjustment instruction is used to adjust the filter intensity of the ocean circulation feedback filter model. The early warning threshold adjustment instruction is used to adjust the early warning threshold of the equipment life cycle assessment index.

[0033] Among them, a multi-dimensional sensor array refers to a combination of sensors installed in different parts of marine equipment to monitor its operating status. This includes accelerometers for detecting mechanical vibrations, temperature sensors for monitoring environmental and equipment temperature changes, strain sensors for measuring structural stress distribution, and electrochemical sensors for assessing seawater corrosion. Ocean circulation intensity refers to the intensity level of water circulation flow in the ocean circulation system, quantified and graded based on a comprehensive evaluation of factors such as ocean circulation velocity, flow direction stability, and eddy intensity. The equipment life cycle assessment index is a quantitative evaluation index that comprehensively considers the current operating status of the equipment, historical degradation trends, and predicted performance degradation. Its value ranges from 0 to 1, with higher values ​​indicating better equipment condition. Ocean circulation velocity refers to the speed of water flow in the ocean, measured using a current meter or acoustic Doppler current profiler. Ocean temperature variation amplitude refers to the maximum change in ocean temperature within a specified time period, calculated through continuous monitoring by temperature sensors. Equipment vibration benchmark value refers to the standard vibration level of marine equipment under normal operating conditions, determined through statistical analysis of vibration data during normal equipment operation. The ocean current circulation interference intensity value is a quantitative indicator of the degree of interference caused by ocean current circulation to monitoring signals, with a value range of 0 to 1. The monitoring system performance evaluation index is a comprehensive indicator for evaluating the overall working effect and data quality of the monitoring system, with a value range of 0 to 1. The parameter adjustment demand coefficient is a quantitative indicator for determining the necessity and magnitude of system parameter adjustments based on the current monitoring status.

[0034] Critical components refer to the core structures and functional areas of marine equipment that play a decisive role in the overall performance, safety, and reliability. Failures in these components will directly affect the normal operation of the equipment and may even lead to system failure. These critical components mainly fall into four categories: the main bearing, a core component supporting rotating parts, plays a crucial role in transmitting loads, reducing friction, and maintaining rotational accuracy; the heat dissipation area is the concentrated area of ​​heat generation and a critical site for heat dissipation, with temperature control being essential for equipment performance and lifespan; the load-bearing structure, structural components that bear the main mechanical loads and resist external forces, is the foundation for ensuring the structural integrity of the equipment; and the seawater contact surface, the surface directly exposed to the marine environment and subjected to seawater corrosion, is a key area for corrosion protection of marine equipment.

[0035] These key components manifest differently in different types of marine equipment. For example, the main bearing components of an offshore wind turbine include the main shaft bearing, yaw bearing, and pitch bearing. The heat dissipation area covers the generator stator windings, gearbox interior, and inverter cooling system. The load-bearing structure consists of the tower structure, foundation connection flanges, and nacelle frame. The seawater contact surfaces are the offshore foundation structure and the anti-corrosion coating at the bottom of the tower. The main bearing components of an offshore oil drilling platform include the drilling rig slewing bearing, mud pump bearing, and crane slewing bearing. The heat dissipation area includes the generator room cooling system, compressor radiator, and hydraulic system cooler. The load-bearing structure consists of the platform main structure, leg connection nodes, and drilling tower. The seawater contact surfaces include the anti-corrosion coating on the platform legs and the inner wall of the ballast tank. Key components of marine engineering vessels include main bearings such as stern tube bearings and main engine crankshaft bearings, heat dissipation areas such as main engine cooling systems and generator radiators, load-bearing structures such as hull main beam structures and deck load-bearing structures, as well as seawater contact surfaces such as the bottom outer plating and propeller surfaces. By arranging appropriate sensors in these key components, accurate monitoring of the equipment's status throughout its entire life cycle can be achieved.

[0036] The specific implementation methods of the above steps are described in detail below.

[0037] The specific implementation of step S01 involves achieving comprehensive condition monitoring of marine equipment through the systematic deployment of a multi-dimensional sensor array. First, sensor installation is planned at key locations on the marine equipment, including the main bearing housing, gearbox, motor housing, radiator surface, main load-bearing beam, connecting flange, deck contact surface, and seawater inlet pipe. A triaxial accelerometer is used as the vibration acceleration sensor, mounted on the surface of the main bearing housing and secured with bolts for a rigid connection. The sensor sensitivity is set to 100mV / g, with a measurement range of ±50g, accurately capturing vibration characteristic signals during bearing operation. A platinum resistance thermometer is used as the temperature sensor, positioned on the radiator surface and motor housing, and tightly bonded to the measured surface with thermally conductive adhesive. The measurement accuracy is ±0.1℃, and the response time is less than 3 seconds, enabling precise monitoring of the equipment's thermal state. Resistance strain gauges are used as stress strain sensors, adhered to high stress concentration areas on the main load-bearing beam and connecting flange. The strain gauge grid length is 3mm, the resistance is 350Ω, and the sensitivity coefficient is 2.1. The strain signal is converted into a voltage signal via a Wheatstone bridge circuit. The seawater corrosion monitoring sensor uses a three-electrode system consisting of a reference electrode and a working electrode. The working electrode is made of the same alloy as the equipment material, the reference electrode is a silver chloride electrode, and the auxiliary electrode is a platinum wire. The distance between the electrodes is controlled within 10 mm. The corrosion state is assessed by measuring the potential change of the working electrode relative to the reference electrode.

[0038] The specific implementation of step S02 involves establishing a high-precision multi-source data acquisition system to achieve synchronous monitoring. The data acquisition system adopts a distributed acquisition architecture, with each sensor node connected to the central data acquisition unit via a fieldbus. Vibration acceleration signal acquisition uses synchronous sampling technology, with three channels simultaneously sampling the X, Y, and Z axes acceleration at a frequency of 1000Hz. A 24-bit analog-to-digital converter ensures signal accuracy, and the signal is digitized after processing by an anti-aliasing filter. Temperature data acquisition uses a cyclic scanning method, measuring each temperature sensor once per second. A 4-wire measurement method is used to eliminate the influence of lead resistance, and the measured data undergoes cold junction compensation and linearization to obtain accurate temperature values. Stress and strain data acquisition is achieved using a strain gauge, with a sampling frequency set to 100Hz. A half-bridge connection is used to improve measurement accuracy. The strain signal is amplified and filtered before being converted into stress values, with the conversion coefficient determined based on the material's elastic modulus. Corrosion potential data acquisition uses an electrochemical workstation, measuring the working electrode potential at a frequency of 0.1Hz with a measurement accuracy of ±1mV. The electrolyte temperature is kept stable during data acquisition. Time synchronization processing uses the network time protocol to ensure that the time error of each acquisition node is less than 1ms. Data preprocessing includes outlier detection, noise filtering and data format unification.

[0039] The specific implementation of step S03 involves using a marine circulation cyclic feedback filtering model to eliminate environmental interference. This model is based on a variational autoencoder architecture to construct a deep filtering network. The encoder part employs a multi-layer convolutional neural network structure to map multi-dimensional monitoring signals from the temporal domain to a low-dimensional latent space. The convolutional kernel size is set to 3×3, the stride is 1, and the activation function uses a modified linear unit function. The latent space processing layer utilizes variational inference principles and employs reparameterization techniques to sample latent variables. The sampling randomness scale parameter is dynamically adjusted according to the intensity of the marine circulation. When the marine circulation cyclic interference coefficient exceeds 0.6, the randomness scale is increased to enhance denoising capabilities. The decoder uses a transposed convolutional structure to reconstruct the clean signal after interference removal. The number of network layers is symmetrically set to the encoder. The adaptive adjustment mechanism of filtering parameters selects different processing strategies based on the interference coefficient. Light filtering uses a Butterworth low-pass filter with a cutoff frequency set to 10 times the signal frequency. Standard filtering adds a median filter stage to eliminate impulse noise. Wavelet denoising processing uses multi-resolution wavelet decomposition, with the wavelet basis function selected being the Dobermann wavelet, and the decomposition layer set to 6 layers.

[0040] The specific implementation of step S04 involves obtaining equipment state characteristic parameters through a multi-domain feature extraction algorithm. The vibration acceleration amplitude change rate is extracted using a sliding window statistical analysis method. The time window length is set to 10 seconds, and the window overlap rate is 50%. Within each time window, the root mean square value of the vibration acceleration is calculated. The change rate is obtained by dividing the difference between the root mean square values ​​of adjacent windows by the time interval. This parameter reflects the dynamic change trend of the equipment's mechanical state. The temperature gradient change rate is calculated using a numerical differential method. A first-order forward differencing operation is performed on the temperature time series, with the differencing step size being the sampling interval. Smoothing filtering is used to eliminate noise amplification effects during the calculation process. This parameter characterizes the rate of change of the equipment's thermal state. The stress concentration factor is extracted through spatial statistical analysis. A spatial stress field is constructed using multi-point stress measurement data. The ratio of the local maximum stress to the regional average stress is calculated. A ratio greater than 2.0 indicates significant stress concentration. This parameter assesses the uniformity of the load distribution on the equipment structure. Corrosion rate calculation employed linear regression analysis, performing least-squares fitting on corrosion potential changes over a specific time period. The slope of the fitted line represented the potential change rate. Faraday's law was used to convert the potential change into corrosion depth rate, with the conversion coefficient determined based on the material's electrochemical equivalent. Eigenvector normalization utilized a min-max standardization method, mapping each characteristic parameter to the 0-1 interval to eliminate the influence of dimensional differences on subsequent analysis.

[0041] The specific implementation of step S05 involves constructing a degradation trend prediction model based on temporal deep learning. This model employs a Long Short-Term Memory (LSTM) network architecture to analyze the temporal evolution of device state characteristics. The network input is a normalized feature vector time series, with a sequence length of 100 time steps, corresponding to approximately one hour of monitoring data. The LTM unit includes three gating mechanisms: a forget gate, an input gate, and an output gate. These gates control the flow of information; the forget gate determines which historical information is discarded, the input gate determines which new information is updated, and the output gate controls the output state. The number of hidden layer neurons is set to 128, employing a bidirectional structure that utilizes both forward and backward temporal information to improve prediction accuracy. The training process uses a time-series cross-validation method, dividing the historical data into a training set and a validation set in chronological order, with the training set accounting for 80% and the validation set for 20%. The loss function is mean squared error, and the optimization algorithm uses adaptive moment estimation. The learning rate is set to 0.001, and the batch size is 64. The model output includes the equipment performance degradation trend curve and the failure probability distribution over the next 24 hours. The probability calculation uses the logistic regression method to map the network output to a probability range of 0 to 1.

[0042] The specific implementation of step S06 involves establishing an equipment lifecycle assessment index calculation and graded response mechanism. The lifecycle assessment index is obtained by weighted fusion of current state parameters, historical degradation trends, and future prediction results, with the current state weighted at 0.4, historical trends at 0.3, and prediction results at 0.3. The current state assessment is based on the deviation of real-time monitoring parameters from normal baseline values; the greater the deviation, the lower the state assessment value. The historical degradation trend assessment calculates the slope of change of each state parameter over the past month; a negative slope indicates performance degradation. The future prediction result assessment is based on the failure probability and performance degradation magnitude output by the prediction model. After the assessment index is calculated, graded response processing is performed. When the index is between 0.8 and 1.0, the equipment is in good condition, and an enhanced monitoring mode is activated, increasing the data acquisition frequency to twice the normal frequency, increasing data storage and transmission bandwidth. When the index is between 0.5 and 0.8, the equipment exhibits slight degradation, triggering a pre-maintenance alarm to notify maintenance personnel to formulate a maintenance plan. The alarm employs a multi-level notification mechanism including email, SMS, and system push notifications. When the index is between 0.2 and 0.5, equipment degradation is significant, and the maintenance plan formulation process is initiated. The system automatically generates a maintenance recommendation list, including inspection items, replacement parts, and maintenance cycles. When the index is below 0.2, the equipment is on the verge of failure, and an equipment replacement recommendation is issued, which includes a replacement time window, a spare parts preparation list, and downtime arrangements.

[0043] The specific implementation of step S07 involves constructing a three-dimensional visualization monitoring interface to intuitively display the equipment status. The visualization model employs three-dimensional graphics rendering technology. The equipment geometric model is established using computer-aided design software, achieving millimeter-level precision and including the main structural components and sensor installation locations. Color mapping uses a heatmap method, mapping monitoring parameter values ​​to a predefined color space. Temperature data uses a red-blue spectrum, stress data uses a green-yellow-red gradient spectrum, vibration data uses a purple-blue-green gradient spectrum, and corrosion data uses a gray-white-red gradient spectrum. Dynamic charts include time-series curves, spectrum analysis charts, and statistical distribution charts. The chart update frequency is synchronized with the data acquisition frequency, supporting historical data playback and trend analysis functions. The user interface supports rotating, zooming, and viewing cross-sections of the three-dimensional model, with mouse operations enabling perspective switching and local magnification. The ocean current circulation feedback control algorithm is integrated through the monitoring interface, displaying the control parameter adjustment status and system performance indicators in real time. Adjustment commands are sent to each acquisition node via network protocols, ensuring dynamic optimization of monitoring system parameters.

[0044] It should be explained that the ocean circulation cyclic feedback filtering model is constructed using a variational autoencoder architecture, with the overall structure comprising three core components: an encoder network, a latent space processing layer, and a decoder network. The encoder network employs a multi-layer convolutional neural network structure. The first convolutional layer contains 32 kernels (1×7) to extract local temporal features. The second convolutional layer contains 64 kernels (1×5) to further abstract feature representations. The third convolutional layer contains 128 kernels (1×3) to obtain high-level semantic features. Each convolutional layer is followed by a batch normalization layer and a modified linear unit activation layer, and downsampling is performed using a max-pooling layer with a 1×2 kernel. The encoder output is mapped to the latent space through a fully connected layer. The latent space has 64 dimensions and includes two branches: a mean vector and a variance vector.

[0045] The latent space processing layer is the core innovation of the variational autoencoder. It achieves latent variable sampling through reparameterization techniques, adhering to the Gaussian distribution assumption during the sampling process. The sampling randomness scale parameter is dynamically determined based on the ocean circulation cycle adjustment function. This function comprehensively considers three environmental parameters: ocean circulation velocity, ocean temperature variation amplitude, and equipment vibration reference value. The ocean circulation cycle interference intensity value is calculated through weighted averaging, with weighting coefficients of 0.4, 0.3, and 0.3. When the interference intensity value falls within different ranges, the latent space sampling randomness scale coefficient is adjusted accordingly: under low interference conditions, the coefficient is set to 0.1 to preserve signal details; under medium interference conditions, the coefficient is set to 0.3 to balance denoising and fidelity; and under high interference conditions, the coefficient is set to 0.8 to enhance denoising capabilities.

[0046] The decoder network structure is symmetrically set with the encoder, employing transposed convolutional layers for upsampling reconstruction. The first transposed convolutional layer contains 128 kernels with a kernel size of 1×3, the second transposed convolutional layer contains 64 kernels with a kernel size of 1×5, and the third transposed convolutional layer contains 32 kernels with a kernel size of 1×7. The final output layer uses a linear activation function to ensure that the signal amplitude is unrestricted. The decoder reconstructs the clean device state signal after removing environmental interference by learning the inverse mapping relationship from the latent representation to the original signal.

[0047] The training dataset creation process comprises three stages: data collection, annotation, and sample construction. The data collection stage is conducted under various ocean circulation conditions. Weak circulation sea states correspond to calm sea states with current speeds less than 0.5 m / s and temperature variations less than 2°C; moderate circulation sea states correspond to general sea states with current speeds of 0.5 to 1.5 m / s and temperature variations of 2 to 5°C; and strong circulation sea states correspond to severe sea states with current speeds greater than 1.5 m / s and temperature variations greater than 5°C. At least 50 hours of continuous monitoring data are collected under each sea state, covering three states: normal equipment operation, minor malfunctions, and severe malfunctions, ensuring the representativeness and completeness of the dataset.

[0048] Data annotation processing employs a combination of manual annotation and algorithm-assisted methods. First, environmental interference and equipment status components in the signal are identified through spectral analysis and time-frequency analysis. Environmental interference signals mainly manifest as low-frequency fluctuations and random noise, while equipment status signals include characteristics such as periodic vibration, temperature cycling, and stress changes. Annotation experts, based on signal characteristics and sea state records, categorize the monitoring data for each time period into two classes: clean status signals and environmental interference signals. Annotation accuracy is ensured to be above 95% through cross-validation. Standardized operating procedures are established during the annotation process to ensure consistency among different annotation personnel.

[0049] In the sample construction phase, the labeled data is organized into a training sample format. Each training sample contains a raw mixed signal with 1024 sampling points as input and the corresponding clean state signal as the label output. Preprocessing of the sample data includes zero-mean normalization and outlier removal to ensure data quality and model training stability. Finally, a standard training sample set containing 10,000 combinations of different ocean circulation intensities and equipment states is constructed. The samples are divided into training, validation, and test sets in a 7:2:1 ratio. The training set is used for model parameter optimization, the validation set for hyperparameter tuning, and the test set for final performance evaluation.

[0050] It should be noted that the key technical ideas of this invention are mainly reflected in four aspects: multi-dimensional sensor fusion monitoring, adaptive filtering of ocean circulation, deep learning degradation prediction, and three-dimensional visualization.

[0051] Multi-dimensional sensor fusion monitoring technology constructs a comprehensive monitoring network covering multiple physical domains, including mechanical, thermal, structural, and chemical domains, by deploying various types of sensors for vibration, temperature, stress, and corrosion at key parts of equipment. Compared to traditional single-parameter monitoring methods, this technology can comprehensively capture the multi-dimensional characteristics of equipment degradation, avoiding missed fault detection due to insufficient monitoring dimensions. Multi-dimensional information fusion, through cross-validation and complementary confirmation, significantly improves the reliability and accuracy of the monitoring system, providing a more complete and accurate data foundation for equipment condition assessment.

[0052] The ocean circulation adaptive filtering technology constructs a depth filtering network based on a variational autoencoder, dynamically adjusting filtering parameters according to the intensity of ocean circulation to achieve intelligent elimination of environmental interference. Traditional fixed-parameter filtering methods struggle to adapt to the complex changes in the marine environment, easily leading to signal distortion or noise residue. This technology, through a potential spatial sampling randomness adjustment mechanism, automatically optimizes the filtering intensity under different sea conditions, ensuring effective removal of interference signals while maximizing the integrity of equipment status signals, significantly improving the adaptability and robustness of signal processing.

[0053] Deep learning-based degradation prediction technology employs long short-term memory networks to analyze the temporal evolution of equipment condition characteristics, enabling accurate prediction of equipment performance degradation trends and failure probabilities. Compared to traditional prediction methods based on empirical models, deep learning technology can automatically learn complex nonlinear degradation patterns without requiring manual mathematical model building, thus exhibiting greater adaptability. The temporal memory mechanism effectively captures the long-term dependencies in the equipment degradation process, resulting in significantly superior prediction accuracy and time span compared to traditional methods, providing a more reliable basis for equipment maintenance decisions.

[0054] 3D visualization technology presents multidimensional monitoring data in an intuitive way through color mapping and dynamic charts, and combines it with the equipment's 3D geometric model to achieve a spatial representation of status information. Traditional 2D chart display methods are difficult to effectively express the spatial distribution and correlation of multidimensional parameters, making them difficult for users to understand. This technology, through 3D spatial mapping and interactive operation, enables users to intuitively understand the status distribution and change trends of various parts of the equipment, significantly improving the readability of monitoring information and decision-making efficiency.

[0055] The synergistic effect of four key technological approaches forms a complete intelligent monitoring solution, achieving end-to-end optimization from data acquisition, signal processing, status analysis to information display. Multi-dimensional sensor fusion provides high-quality data sources for subsequent processing, adaptive filtering of ocean circulation ensures data purity, deep learning degradation prediction enables intelligent analysis, and 3D visualization enhances user experience. These four technologies support and complement each other, forming a powerful synergy that far surpasses the effect of simply adding up the individual technologies, providing advanced technical means and solutions for the lifecycle management of marine equipment.

[0056] It should be noted that this invention also solves the following technical problem: the inability of existing marine equipment monitoring systems to adaptively adjust monitoring parameters according to environmental changes. Traditional monitoring systems typically employ fixed data acquisition frequencies, preset alarm thresholds, and a single signal processing mode, which cannot adapt to the dynamic characteristics of the marine environment. When sea conditions change, the monitoring system continues to operate in a fixed mode, resulting in insufficient data acquisition under severe sea conditions or wasted resources under calm sea conditions. This invention establishes a cyclic feedback relationship between monitoring parameters through a marine circulation feedback control algorithm. It calculates a monitoring system performance evaluation index based on state characteristic parameters such as the rate of change of vibration acceleration amplitude, the rate of change of temperature gradient, the stress concentration factor, and the corrosion rate. When the performance index deviates from a preset threshold, it automatically calculates the parameter adjustment requirement coefficient and outputs data acquisition frequency adjustment commands, filter intensity adjustment commands, and early warning threshold adjustment commands, thereby achieving intelligent dynamic optimization of monitoring system parameters.

[0057] Furthermore, this invention addresses the technical problem of traditional marine equipment condition monitoring lacking multi-dimensional information fusion and analysis capabilities. Existing monitoring technologies primarily rely on single-type sensors or independent analysis of multiple sensors for condition assessment, failing to fully utilize the correlation and complementarity between different sensor information, resulting in an incomplete understanding of the overall equipment condition. This invention employs a multi-dimensional sensor array collaborative monitoring mechanism, performing time-synchronized processing and data preprocessing on monitoring data of various physical quantities such as vibration, temperature, stress, and corrosion, establishing an equipment condition feature vector matrix to achieve multi-source information fusion analysis. By extracting and normalizing feature parameters from different sensor data, and utilizing time-series deep learning algorithms to uncover deep correlations between multi-dimensional condition parameters, it can comprehensively assess the equipment's operating status and degradation level from multiple dimensions, significantly improving the comprehensiveness and accuracy of condition monitoring and providing more reliable technical support for equipment lifecycle management.

[0058] Specifically, the principle of this invention is as follows: The key to solving the technical problem of low monitoring accuracy caused by environmental interference lies in establishing a deep filtering mechanism capable of intelligently identifying and adaptively filtering out environmental interference. Traditional filtering methods use fixed filtering parameters and preset frequency cutoff points, which cannot adapt to the dynamic changes in the marine environment. The marine circulation cyclic feedback filtering model based on a variational autoencoder architecture used in this invention can automatically learn the essential differences between environmental interference signals and equipment status signals through a deep learning network. The encoder part of this model compresses and maps multidimensional monitoring signals containing environmental interference into a low-dimensional latent space. In the latent space, the randomness of the variational sampling mechanism is used to model the probability distribution of signal features. The degree of separation of different signal components is controlled by adjusting the randomness scale parameter of the latent space sampling. The marine circulation cyclic adjustment function calculates the intensity value of marine circulation cyclic interference by comprehensively analyzing three key environmental parameters: marine circulation velocity, marine temperature variation amplitude, and equipment vibration reference value. Based on this intensity value, the randomness scale coefficient of the latent space sampling is dynamically determined, achieving real-time matching between filtering parameters and environmental conditions. When the intensity of ocean current circulation interference is low, a small random scale coefficient is used for light filtering to retain more detailed equipment status information. When the interference intensity is high, the random scale coefficient is increased to strengthen the filtering intensity and effectively suppress the impact of environmental noise on the monitoring signal. The decoder section reconstructs the equipment status signal after removing environmental interference based on the clean features separated in the potential space. This end-to-end deep filtering process achieves accurate filtering of interference from complex marine environments, ensuring that the extracted equipment status feature parameters can truly reflect the actual operating status and degradation trend of the equipment.

[0059] The following provides a specific embodiment 1 of the present invention, and the specific implementation of each step in this embodiment 1 is described in detail below.

[0060] In this embodiment, the specific implementation of steps S01-S02 is the same as described above, and will not be repeated in detail here.

[0061] The specific implementation of step S03 involves using an ocean circulation feedback filtering model to filter out environmental interference from synchronous multi-source monitoring data. The ocean circulation interference coefficient C... disturb The specific calculation is expressed as follows:

[0062] C disturb =αV current +βΔT ocean +γV vibration_base +ε1;

[0063] In the formula, C disturb V is the ocean current circulation disturbance coefficient, dimensionless, with a numerical range of 0 to 1; current ΔT represents the ocean current velocity, expressed in m / s. ocean V represents the range of ocean temperature variation, expressed in °C. vibration_base The vibration reference value for the equipment is expressed in m / s². 2 α, β, and γ are weighting coefficients, dimensionless, with values ​​of 0.4, 0.3, and 0.3 respectively; ε1 is the error term, dimensionless, ranging from 0.05 to 0.1. The parameters are obtained as follows: V current The current was measured using a current meter, including the following steps: Step 1: Install the current meter on the seabed within 10 meters of the equipment; Step 2: Measure the current velocity vector continuously for 24 hours; Step 3: Calculate the vector magnitude as the circulation velocity value. ΔT ocean Temperature data is obtained using temperature sensors, including step 1: deploying seawater temperature sensors near the equipment; step 2: continuously collecting temperature data for 24 hours; and step 3: calculating the difference between the maximum and minimum temperatures. vibration_base The data was obtained using statistical analysis methods, including step 1: collecting vibration data for the first 100 hours of normal operation of the equipment; step 2: calculating the root mean square value of the vibration data; and step 3: taking the root mean square value as the vibration reference value.

[0064] The specific implementation of step S04 is to extract the state characteristic parameters of marine equipment. Vibration acceleration amplitude change rate R vibration The specific calculation is expressed as follows:

[0065]

[0066] In the formula, R vibration The rate of change of vibration acceleration amplitude, in m / s². 3 RMS i+1 This represents the root mean square value of the vibration acceleration in the (i+1)th time window, in m / s². 2 RMSi The root mean square value of the vibration acceleration in the i-th time window, in m / s². 2 Δt represents the time interval, which is 10 seconds. The root mean square (RMS) value is also included. i The calculation is expressed as follows:

[0067]

[0068] In the formula, RMS i This represents the root mean square value for the i-th time window, in m / s. 2 N represents the number of sampling points within the time window, which is dimensionless; a j This represents the acceleration value at the j-th sampling point, in m / s². 2 .

[0069] Temperature gradient change rate R temperature The specific calculation is expressed as follows:

[0070]

[0071] In the formula, R temperature T represents the rate of change of temperature gradient, in °C / s. k+1 T represents the temperature value at time point k+1, in °C. k Δt represents the temperature value at the k-th time point, in °C. temp This represents the temperature sampling time interval, with a value of 1 second.

[0072] Stress concentration factor K stress The specific calculation is expressed as follows:

[0073]

[0074] In the formula, K stress σ is the stress concentration factor, dimensionless; max The maximum stress value within the measurement area, in Pa; σ avg The average stress value within the measurement area is expressed in Pa; δ is a correction term, dimensionless, with a value ranging from 0.05 to 0.1.

[0075] Wherein, the average stress value σ avg The calculation is expressed as follows:

[0076]

[0077] In the formula, σ avg The average stress value is expressed in Pa; M is the number of stress measurement points, which is dimensionless; σ m This represents the stress value at the m-th measurement point, in Pa.

[0078] Corrosion rate V corrosionThe specific calculation is expressed as follows:

[0079]

[0080] In the formula, V corrosion , where is the corrosion rate in mm / year; k is the potential conversion coefficient in mm·year / (mV), determined based on the material's electrochemical equivalent, ranging from 0.001 to 0.01. ε0 represents the slope of the corrosion potential change over time, in mV / s; ε2 is the error term, in mm / year, ranging from 0.02 to 0.05. The slope of the corrosion potential change was obtained through linear fitting analysis, as detailed below:

[0081]

[0082] In the formula, The slope of the corrosion potential change is expressed in mV / s; P is the number of measurement points, dimensionless, typically with no fewer than 100 points; t p For the p-th measurement time point, the unit is seconds; E p This represents the corrosion potential value at the p-th measurement time point, in mV. The average value over time, expressed in seconds, is calculated using the following formula: The average potential is expressed in mV and is calculated using the following formula:

[0083] Device state feature vector matrix X feature The representation is as follows:

[0084]

[0085] In the formula, X feature This represents the device status feature vector matrix; n is the length of the time series; the subscript numbers represent the feature values ​​at different time points. Normalization is performed using the min-max normalization method, as shown below:

[0086]

[0087] In the formula, X norm X is the normalized eigenvalue matrix; min X is the minimum value in the characteristic matrix; max It represents the maximum value in the characteristic matrix.

[0088] The specific implementation method of step S05 is the same as described above, and will not be repeated in detail here.

[0089] The specific implementation of step S06 involves calculating the equipment lifecycle assessment index based on the equipment degradation prediction results. Equipment Lifecycle Assessment Index I lifecycle The specific calculation is expressed as follows:

[0090] I lifecycle =w1S current +w2T history +w3P future +ε3;

[0091] In the formula, I lifecycle S is an equipment life cycle assessment index, dimensionless, with a value range of 0 to 1. current The current state assessment value is dimensionless and ranges from 0 to 1; T history This is a dimensionless assessment value for historical degradation trends, ranging from 0 to 1; P future The evaluation value for the future prediction result is dimensionless, ranging from 0 to 1; w1, w2, and w3 are weighting coefficients, dimensionless, with values ​​of 0.4, 0.3, and 0.3 respectively; ε3 is the error term, dimensionless, ranging from 0.01 to 0.05. The current state evaluation value S... current The calculation is expressed as follows:

[0092]

[0093] In the formula, S current X is the current state evaluation value, dimensionless; current The current feature vector, after normalization, has dimensionless components; X baseline The baseline eigenvector, after normalization, has dimensionless components, representing the baseline characteristic value under normal operating conditions; |||| represents the Euclidean norm. Historical degradation trend assessment value T. history The calculation is expressed as follows:

[0094]

[0095] In the formula, T history ΔX is a dimensionless value representing the historical degradation trend assessment; Q is the number of historical time periods, also dimensionless, typically 30 representing the past 30 days; ΔX q Let be the characteristic change in the q-th time period, which is dimensionless, and the calculation formula is: Where X q Let be the mean of the feature vectors in the q-th time period. The future prediction result evaluation value P. future The calculation is expressed as follows:

[0096] P future =1-P failure ;

[0097] In the formula, P future The value is an evaluation of the future prediction results, dimensionless; P failureTo predict the probability of failure, a dimensionless value ranging from 0 to 1 is output by a deep learning prediction model.

[0098] The specific implementation of step S07 involves constructing a three-dimensional visualization model of the marine equipment lifecycle. The performance evaluation index P of the monitoring system in the ocean circulation feedback control algorithm... system The specific calculation is expressed as follows:

[0099] P system =w4R vibration +w5R temperature +w6K stress +w7V corrosion +ε4;

[0100] In the formula, P system The system performance evaluation index is dimensionless, ranging from 0 to 1; w4, w5, w6, and w7 are weighting coefficients, dimensionless, with values ​​of 0.3, 0.2, 0.3, and 0.2 respectively; ε4 is the error term, dimensionless, ranging from 0.02 to 0.08. To ensure dimensional uniformity, all input parameters need to be normalized, as shown below:

[0101]

[0102] In the formula, the subscript `norm` represents the normalized value, and the subscript `max` represents the maximum reference value for each parameter. The parameter adjustment demand coefficient D... adjust The calculation is expressed as follows:

[0103] D adjust =|P system -P thrshold |·λ;

[0104] In the formula, D adjust The parameter adjustment demand coefficient is dimensionless; P threshold λ is the preset performance threshold, dimensionless, with a value of 0.8; λ is the adjustment sensitivity coefficient, dimensionless, with a value of 1.2.

[0105] Ocean circulation regulation function F regulation The specific calculation is expressed as follows:

[0106] F regulation =μ1V current +μ2ΔT ocean +μ3V vibration_base ;

[0107] In the formula, F regulationThe value represents the ocean current circulation regulation function, dimensionless, ranging from 0 to 1; μ1, μ2, and μ3 are weighting coefficients, dimensionless, with values ​​of 0.4, 0.3, and 0.3 respectively. To ensure dimensional uniformity, all input parameters need to be normalized, as shown below:

[0108]

[0109] In the formula, V current , max The maximum reference value for ocean circulation velocity is 3.0 m / s; ΔT ocean , max The maximum reference value for ocean temperature variation is 10℃; V vibration_base，max The maximum reference value for equipment vibration is set at 10 m / s². 2 The scale parameter σ of the latent spatial sampling randomness sampling Determined based on the adjustment function value, the specific relationship is expressed as follows:

[0110]

[0111] In the formula, σ sampling The scale parameter for the randomness of sampling in the latent space is dimensionless.

[0112] It should be explained that the formula for calculating the ocean circulation interference coefficient is based on the principle of multi-parameter weighted fusion. By comprehensively considering three key environmental factors—ocean circulation velocity, temperature change amplitude, and equipment vibration reference value—it achieves a quantitative assessment of the intensity of ocean environmental interference.

[0113] C disturb =αV current +βΔT ocean +γV vibration_base +ε1;

[0114] Compared with traditional single-parameter evaluation methods, this formula can more comprehensively and accurately reflect the combined impact of complex marine environments on monitoring signals, providing a precise basis for parameter adjustment in subsequent adaptive filtering processing, and significantly improving the environmental adaptability and signal processing accuracy of the filtering system.

[0115] The formula for calculating the rate of change of vibration acceleration amplitude is based on the principle of time-domain difference analysis. It characterizes the dynamic change trend of the equipment's mechanical state by calculating the rate of change of the root mean square value of adjacent time windows.

[0116]

[0117] This formula, combined with the root mean square value calculation formula, can effectively extract the energy change characteristics of vibration signals.

[0118]

[0119] Compared with traditional peak detection methods, it has stronger anti-noise interference capability and higher sensitivity to state changes, providing a reliable quantitative indicator for early diagnosis of equipment mechanical faults.

[0120] The corrosion rate calculation formula is based on the principle of electrochemical reaction kinetics. It quantitatively assesses the degree of material corrosion by analyzing the trend of corrosion potential changes through linear fitting.

[0121]

[0122] The slope of the potential change was obtained by fitting using the least squares method.

[0123]

[0124] Compared with the traditional weight loss method, this formula has higher measurement accuracy and real-time performance, and can detect corrosion acceleration phenomena in a timely manner, providing a scientific basis for the corrosion prevention and maintenance of marine equipment.

[0125] The equipment life cycle assessment index calculation formula is based on the principle of multi-time series information fusion. It achieves a comprehensive quantitative evaluation of the equipment life cycle status by weightedly combining the assessment results of three dimensions: current status, historical trends, and future predictions.

[0126] I lifecycle =w1S current +w2T history +w3P future +ε3;

[0127] The current state assessment uses Euclidean distance to measure the degree of deviation.

[0128]

[0129] Compared to traditional single-point-in-time status assessment methods, this formula can more comprehensively reflect the overall health status and development trend of equipment, providing a more scientific and reasonable basis for maintenance decisions and effectively improving the foresight and accuracy of equipment management.

[0130] The performance evaluation index calculation formula of the monitoring system is based on the principle of multi-domain feature fusion. It achieves a comprehensive evaluation of the overall performance of the monitoring system by weighted combination of state parameters of four physical domains: vibration, temperature, stress, and corrosion.

[0131] P system =w4R vibration +w5R temperature +w6K stress +w7V corrosion +ε4;

[0132] This formula, combined with the calculation of parameter adjustment demand coefficients, can automatically determine the necessity and adjustment range of parameter adjustment based on the degree of deviation of system performance.

[0133] D adjust =|P system -P threshold |·λ;

[0134] Compared to traditional fixed threshold triggering methods, it has stronger adaptability and more precise control, significantly improving the intelligence level and operating efficiency of the monitoring system.

[0135] The ocean circulation regulation function is based on the principle of environmental parameter mapping, and determines the regulation intensity of the filtering model by linearly combining key ocean environmental parameters.

[0136] F regulation =μ1V current +μ2ΔT ocean +μ3V vibration_base ;

[0137] This function, combined with segmented potential space sampling parameter settings, can automatically adjust the processing intensity of the filtering algorithm according to different marine environmental conditions.

[0138]

[0139] Compared with traditional fixed-parameter filtering methods, it has stronger environmental adaptability and higher signal fidelity, effectively solving the technical problem of balancing signal processing accuracy and noise reduction effect in complex marine environments.

[0140] To better understand and implement this invention, the following is a specific application scenario example 2: A technical team needs to conduct full life-cycle status monitoring of wind turbine generators operating in a complex marine environment. This wind farm is constantly affected by monsoons and ocean currents, with ocean circulation speeds varying from 0.5 to 2.8 m / s and seawater temperature fluctuating seasonally by up to 18°C, causing severe interference to equipment monitoring signals. The technical team decided to use an artificial intelligence-based visual monitoring method for the life-cycle of marine equipment to solve this technical challenge.

[0141] The technical team first deployed a multi-dimensional sensor array at key locations on five 3MW offshore wind turbine generators. Three vibration acceleration sensors were installed at the main bearing to monitor vibration signals in the X, Y, and Z directions, with a sampling frequency of 1000Hz. Eight temperature sensors were installed in the generator cooling system, with a sampling frequency of 1Hz, to monitor bearing temperature rise and heat dissipation efficiency. Twelve stress-strain sensors were deployed at the load-bearing structure at the connection between the tower and nacelle, with a sampling frequency of 100Hz, to monitor structural stress distribution in real time. Six seawater corrosion monitoring sensors were installed at the tower foundation, where it is in direct contact with seawater, with a sampling frequency of 0.1Hz, to monitor changes in electrochemical corrosion potential.

[0142] During the 30-day continuous monitoring period, the system collected a large amount of multi-source monitoring data. Vibration acceleration signals showed that the root mean square value of vibration at the main bearing location ranged from 2.1 to 8.7 m / s². 2 The vibration fluctuated within a certain range, with the maximum amplitude reaching 12.3 m / s in the X-axis direction. 2 Temperature monitoring data shows that the temperature in the heat dissipation area varies between 35 and 78℃, with a maximum temperature gradient rate of 2.4℃ / min. Stress distribution data shows that the maximum stress value of the load-bearing structure is 156 MPa, the average stress value is 89 MPa, and the stress concentration factor is 1.75. Corrosion potential monitoring results show that the corrosion potential of the seawater contact surface varies within the range of -0.65 to -0.42V, with a corrosion rate of 0.083 mm / year.

[0143] The technical team established a training dataset for the ocean circulation feedback filtering model and collected monitoring data under different ocean circulation conditions. Table 1 shows the circulation parameters under three typical sea states:

[0144] Table 1 Ocean circulation parameters under different sea states

[0145]

[0146] The technical team manually labeled the data for each operating condition, establishing a standard training sample set containing 10,000 combinations of different ocean circulation intensities and equipment states. A supervised learning approach was used for end-to-end training of the variational autoencoder. The loss function incorporated reconstruction error and KL divergence regularization terms, with a weight ratio of 0.7:0.3. An adaptive learning rate scheduling strategy was employed during training. The initial learning rate was set to 0.001, and it decayed to 0.8 times its original value every 50 training epochs, for a total of 500 training epochs. A stochastic gradient descent optimizer with a batch size of 32 was used.

[0147] During the environmental interference filtering stage, the system automatically adjusts the filtering parameters based on the intensity of ocean current circulation. When the ocean current circulation interference coefficient is 0.73, wavelet denoising is used, and the latent space sampling randomness scaling factor is set to 0.8. When the interference coefficient is 0.48, standard filtering is used, and the scaling factor is set to 0.3. When the interference coefficient is 0.21, light filtering is used, and the scaling factor is set to 0.1. The signal quality is significantly improved after filtering, with noise amplitude reduced by 68% and effective signal fidelity reaching 96.2%.

[0148] The technical team extracted the state characteristic parameters of the marine equipment and established a state characteristic vector matrix. The rate of change of vibration acceleration amplitude was calculated through time-domain statistical analysis, with a continuous time window of 10 seconds and an average rate of change of 0.134 seconds. -1 The temperature gradient change rate was obtained through first-order difference calculation, with an interval of 1 second between adjacent sampling points and an average change rate of 1.67℃ / min. The stress concentration factor was calculated through spatial statistical analysis, with a maximum value to average value ratio of 1.75. The corrosion rate was obtained through linear fitting analysis, with a corrosion potential change slope of -0.018V / month, which translates to a corrosion depth rate of 0.083mm / year.

[0149] Table 2 shows the extracted key equipment status feature parameters:

[0150] Table 2 Results of Equipment Status Feature Parameter Extraction

[0151]

[0152] The system establishes a model for predicting the degradation trend of marine equipment, employing a Long Short-Term Memory (LSTM) network as the core of the temporal deep learning algorithm. The network structure consists of three LSTM layers with 128 neurons per layer, and a dropout rate of 0.2 is set to prevent overfitting. The model input is 30 days of historical feature data, predicting the performance degradation trend over the next 15 days. The prediction results show that the main bearing vibration trend will increase over the next 7 days, with a predicted vibration amplitude reaching 9.2 m / s. 2 The probability of failure is 0.18.

[0153] Based on the equipment degradation prediction results, the system calculated the equipment life cycle assessment index. Taking into account the current operating status, historical degradation trends, and predicted performance decline, the life cycle assessment indices for the five units were 0.76, 0.82, 0.69, 0.73, and 0.78, respectively. Table 3 shows the monitoring strategies corresponding to different assessment index ranges:

[0154] Table 3 Comparison of Equipment Life Cycle Assessment Index and Monitoring Strategy

[0155] [0.8,1.0) Strengthen monitoring mode 50% frequency increase Normal operation Regular inspection [0.5,0.8) Pre-maintenance alarm 30% frequency increase Yellow Alert Planned maintenance [0.2,0.5) Maintenance plan development Frequency increased by 80% Orange alert Real-time maintenance [0,0.2) Equipment replacement recommendations 100% frequency increase Red Alert Shutdown and maintenance

[0156] The system constructs a 3D visualization model of the marine equipment lifecycle, displaying multi-source monitoring data in real time using color mapping. Vibration data is represented by a blue gradient, temperature data by a red gradient, stress data by a green gradient, and corrosion data by a yellow gradient. Dynamic charts use a time axis on the horizontal axis and characteristic parameter values ​​on the vertical axis, updating in real time to show the trend of equipment status changes. The visualization interface integrates the 3D model of the equipment, with different parts changing colors in real time according to monitoring data, intuitively reflecting the health status of the equipment.

[0157] The ocean circulation feedback control algorithm automatically optimizes system performance by establishing a cyclic feedback relationship between monitoring parameters. The parameter input module receives four key characteristic parameters, and the cyclic feedback calculation module calculates the monitoring system performance evaluation index using a weighted summation algorithm, with vibration parameters having a weight of 0.3, temperature parameters 0.2, stress parameters 0.3, and corrosion parameters 0.2. When the performance evaluation index is 0.67, the system automatically outputs a data acquisition frequency adjustment command, increasing the vibration sensor sampling frequency from 1000Hz to 1300Hz. When the performance evaluation index drops to 0.54, the system outputs a filter intensity adjustment command, adjusting the potential space sampling randomness scale coefficient from 0.3 to 0.5.

[0158] Table 4 shows the performance statistics after the system has been running for 30 days:

[0159] Table 4 System Performance Statistics Results

[0160] Signal filtering accuracy 96.2 ％ 82.1 17.2％ Fault prediction accuracy 94.8 ％ 79.3 19.5％ False alarm rate 3.7 ％ 12.4 Reduced by 70.2% Data processing delay 125 ms 340 Reduced by 63.2% System Availability 99.1 ％ 94.6 4.8％

[0161] Compared to traditional single-sensor monitoring methods, this invention employs a multi-dimensional sensor array and artificial intelligence filtering algorithms, significantly improving monitoring accuracy in marine environments. Traditional methods achieve a signal-to-noise ratio (SNR) of only 15.6 dB under strong ocean current interference, while this invention achieves 23.8 dB, a 52% improvement. Regarding fault prediction, traditional threshold-based methods have an accuracy of 79.3%, while this invention's deep learning-based prediction model achieves 94.8%, a 19.5% improvement. Traditional monitoring systems cannot effectively distinguish between environmental interference and equipment anomalies, resulting in a false alarm rate as high as 12.4%. This invention reduces the false alarm rate to 3.7% through an ocean current cyclic feedback filtering model. In terms of real-time data processing, traditional methods require 340 ms to complete a full data processing cycle, while this invention shortens the processing time to 125 ms through optimized algorithm structure. Most importantly, this invention enables a visualized display of the equipment's entire lifecycle status, providing technicians with intuitive equipment health status information, whereas traditional methods only provide numerical monitoring results and cannot form a systematic status assessment.

[0162] It should be noted that the variables involved in this invention are explained in detail in Tables 5 and 6 below.

[0163] Table 5. Variable Explanation Table (Part 1)

[0164]

[0165] Table 6. Variable Explanation Table (Part Two)

[0166]

[0167]

[0168] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for visual monitoring of the lifecycle of marine equipment based on artificial intelligence, characterized in that, include: Deploy multi-dimensional sensor arrays on marine equipment; Multi-source monitoring data is collected during the operation of marine equipment and time synchronization and data preprocessing are performed to obtain synchronous multi-source monitoring data. Environmental interference is filtered out from synchronous multi-source monitoring data using an ocean circulation feedback filtering model; marine equipment status characteristic parameters are extracted, and an equipment status characteristic vector matrix is ​​established and normalized; a marine equipment degradation trend prediction model is established, and the evolution law of the equipment status characteristic vector matrix is ​​analyzed using a time-series deep learning algorithm to predict the performance degradation trend and failure probability of the equipment in the future time period, and output the equipment degradation prediction results; the equipment life cycle assessment index is calculated based on the equipment degradation prediction results, and corresponding monitoring strategies are executed according to the index range; a three-dimensional visualization model of the marine equipment life cycle is constructed, which displays synchronous multi-source monitoring data, equipment status characteristic vector matrix, equipment degradation prediction results, and equipment life cycle assessment index in real time in the form of color mapping and dynamic charts, forming an intuitive display interface of the equipment's full life cycle status, and the monitoring system parameters are dynamically adjusted through an ocean circulation feedback control algorithm; Among them, the ocean circulation feedback filtering model is a deep filtering network based on the variational autoencoder architecture, which includes an encoder layer, a latent space processing layer and a decoder layer. The encoder maps multidimensional monitoring signals to a low-dimensional latent space. The latent space processing layer uses the sampling randomness scale parameter of the latent space to adaptively separate the signal features. The decoder reconstructs the pure equipment status signal after removing environmental interference. Among them, the potential space sampling random scale parameter is dynamically determined by the ocean circulation cycle adjustment function based on three parameters: ocean circulation velocity, ocean temperature variation amplitude, and equipment vibration reference value. Specifically, the ocean circulation velocity is obtained by measuring the current velocity vector of the water flow at the current location through a current meter installed near the marine equipment and calculating its modulus. The ocean temperature variation amplitude is obtained by analyzing the temperature data of the temperature sensor over the past 24 hours and calculating the difference between the maximum and minimum temperatures. The equipment vibration reference value is obtained by statistically analyzing the vibration data of the vibration acceleration sensor during the first 100 hours of normal operation of the equipment and calculating the root mean square value.

2. The method for visual monitoring of the life cycle of marine equipment according to claim 1, characterized in that, The steps of the multi-dimensional sensor array are as follows: Vibration acceleration sensors, temperature sensors, stress and strain sensors, and seawater corrosion monitoring sensors are deployed in key parts of marine equipment. Vibration acceleration sensors are deployed in the main bearings of the equipment, temperature sensors are deployed in the heat dissipation areas of the equipment, stress and strain sensors are deployed in the load-bearing structures of the equipment, and seawater corrosion monitoring sensors are deployed on the seawater contact surfaces of the equipment, forming a multi-parameter monitoring network covering the entire life cycle of the equipment.

3. The method for visual monitoring of the life cycle of marine equipment according to claim 2, characterized in that, The steps for collecting multi-source monitoring data are as follows: vibration acceleration signal is continuously collected from the triaxial acceleration data of the main bearing of the equipment through a vibration acceleration sensor at a sampling frequency of 1000Hz; temperature change curve is continuously monitored from the temperature sensor at a sampling frequency of 1Hz to the temperature data of the heat dissipation area of ​​the equipment; stress distribution data is measured from the stress strain sensor at a sampling frequency of 100Hz to the strain value of the load-bearing structure of the equipment in real time and converted into stress data; and corrosion potential change data is monitored from the seawater corrosion monitoring sensor at a sampling frequency of 0.1Hz to the electrochemical corrosion potential of the seawater contact surface of the equipment.

4. The method for visual monitoring of the life cycle of marine equipment according to claim 3, characterized in that, The steps for filtering environmental interference using the ocean circulation feedback filtering model are as follows: the filtering parameters are automatically adjusted according to the ocean circulation intensity. When the ocean circulation interference coefficient ∈ (0.6, 1.0], wavelet denoising is added; when the ocean circulation interference coefficient ∈ [0.3, 0.6], standard filtering is used; and when the ocean circulation interference coefficient ∈ [0, 0.3), mild filtering is used.

5. The method for visual monitoring of the life cycle of marine equipment according to claim 4, characterized in that, The steps for extracting the state characteristic parameters of marine equipment are as follows: the rate of change of vibration acceleration amplitude is obtained by performing time-domain statistical analysis on the vibration acceleration signal and calculating the rate of change of the root mean square value of vibration acceleration within a continuous time window; the rate of change of temperature gradient is obtained by performing first-order difference operation on the temperature change curve, calculating the difference of temperature values ​​at adjacent time points and then dividing by the time interval; the stress concentration factor is obtained by performing spatial statistical analysis on the stress distribution data and calculating the ratio of the maximum stress value to the average stress value; and the corrosion rate is obtained by performing linear fitting analysis on the corrosion potential change data, calculating the slope of the corrosion potential change with time and converting it into the corrosion depth rate.

6. The method for visual monitoring of the life cycle of marine equipment according to claim 5, characterized in that, The steps for implementing the monitoring strategy based on the equipment life cycle assessment index are as follows: when the equipment life cycle assessment index is in the range of [0.8, 1.0), the enhanced monitoring mode is activated to increase the data collection frequency; when the equipment life cycle assessment index is in the range of [0.5, 0.8), a pre-maintenance alarm is triggered; when the equipment life cycle assessment index is in the range of [0.2, 0.5), the maintenance plan formulation procedure is initiated; and when the equipment life cycle assessment index is in the range of [0, 0.2), an equipment replacement recommendation is issued.

7. The method for visual monitoring of the life cycle of marine equipment according to claim 6, characterized in that, The ocean circulation regulation function is specifically calculated by weighting three parameters—ocean circulation velocity, ocean temperature variation amplitude, and equipment vibration reference value—to obtain the ocean circulation disturbance intensity value. When the ocean circulation disturbance intensity value ∈ [0, 0.3), the potential space sampling randomness scaling coefficient is set to 0.1; when the ocean circulation disturbance intensity value ∈ [0.3, 0.6), the potential space sampling randomness scaling coefficient is set to 0.3; and when the ocean circulation disturbance intensity value ∈ [0.6, 1.0], the potential space sampling randomness scaling coefficient is set to 0.

8.

8. The method for visual monitoring of the life cycle of marine equipment according to claim 7, characterized in that, The ocean circulation feedback control algorithm specifically includes a parameter input module, a feedback calculation module, and a parameter adjustment output module. The parameter input module receives the rate of change of vibration acceleration amplitude, the rate of change of temperature gradient, the stress concentration factor, and the corrosion rate as input parameters. The feedback calculation module calculates the monitoring system performance evaluation index based on the input parameters using a weighted summation algorithm. The parameter adjustment output module outputs data acquisition frequency adjustment commands, filter intensity adjustment commands, and early warning threshold adjustment commands based on the monitoring system performance evaluation index.

Images