Propeller fault diagnosis and predictive maintenance method and system

Abstract

The invention relates to the technical field of data processing, and discloses a propeller fault diagnosis and predictive maintenance method and system. The method comprises the following steps: collecting current, temperature, vibration and pressure signals of a propeller, extracting characteristic parameters to construct a time sequence data set, mapping the characteristic parameters into nodes, establishing a multi-physical field coupling hyperedge to construct a time sequence incidence matrix, inputting an encoder, extracting a degradation state characteristic vector, and mapping the degradation state characteristic vector into a health index; and calculating a degradation rate function in combination with the working condition parameters, and integrating the degradation rate function to predict the remaining service life. According to the method, the high-order coupling relation and the time sequence evolution characteristics among multiple parameters of the propeller are captured through the multi-physics field coupling hypergraph time sequence encoder, the influence of different operation conditions on degradation is quantified in combination with the degradation rate model sensitive to the working condition, and the problem that the fault evolution trajectory and the remaining service life cannot be predicted is solved; and an accurate time window reference is provided for predictive maintenance.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, and in particular to a propeller fault diagnosis and predictive maintenance method and system. BACKGROUND

[0002] Propeller fault diagnosis and predictive maintenance technology is the key to ensure the safe operation of autonomous underwater vehicles. Traditional propeller maintenance mainly relies on periodic maintenance and after-service mode, such as replacing bearings and checking sealing systems at fixed intervals, but this method cannot dynamically adjust the maintenance time according to the actual degradation state and operating conditions of the propeller, resulting in resource waste due to excessive maintenance or sudden failure due to insufficient maintenance. In recent years, data-driven fault diagnosis methods have been gradually applied to propeller health monitoring. By collecting operating parameters such as current, temperature, and vibration, machine learning algorithms are used to identify the current fault type of the propeller, such as using neural networks to classify whether the propeller has bearing wear or sealing leakage, or using support vector machines to judge the health status level of the propeller according to the vibration spectrum characteristics. Some studies introduce graph neural network technology to process the correlation between sensor data, such as a semi-supervised fault diagnosis method based on parallel hypergraph attention network and graph convolution network, which accurately identifies the fault type of the propeller in small samples and interference environments by constructing high-order relationships between nodes and learning dynamic and static features in parallel, solving the problems of insufficient labeled training samples and ocean environment interference.

[0003] However, existing propeller fault diagnosis methods mainly focus on instantaneous state classification, i.e., determining the current fault state of the propeller, but lack the ability to model the evolution process from the inception to the failure of the fault. When the propeller operates underwater, the degradation processes such as magnetic coupling transmission torque attenuation, bearing gap increase, and oil-filled sealing cabin pressure imbalance exhibit gradual evolution characteristics, and are influenced by various factors such as water depth pressure gradient, load fluctuation, and start-stop impact. The degradation rate under different working conditions differs significantly. Although existing static graph network-based diagnosis methods can capture the instantaneous correlation between sensor parameters, they cannot track the dynamic changes of these correlations over time and cannot establish the time evolution rules of coupled degradation between current harmonic distortion, temperature rise rate, vibration characteristics, and pressure response. In addition, key components of the propeller, such as bearings mentioned in the propeller manual, have a 2000-hour life cycle, but the actual remaining life is affected by factors such as deep water high pressure accelerating seal aging, high load accelerating magnetic coupling wear, and frequent start-stop cumulative damage. Simple life prediction based on operating hours cannot reflect the modulation effect of real working conditions on the degradation rate.

[0004] Due to the lack of time evolution modeling in existing methods, it is impossible to infer future degradation trends from current fault diagnosis results, and it is impossible to answer the key questions of "how will the fault develop" and "when will maintenance be needed". Specifically, existing methods do not establish a time sequence correlation model of the coupling relationship between multi-physical field parameters, and cannot capture the time-varying coupling mechanism across physical fields such as motor heating causing oil pressure changes, bearing wear causing vibration characteristics to evolve together, etc. Therefore, it is difficult to identify the evolution path branching characteristics of different degradation modes on the time axis, and it is difficult to distinguish between bearing wear dominant degradation, seal failure dominant degradation, and motor insulation aging dominant degradation. In addition, existing methods do not establish a working condition sensitive degradation rate model, which cannot quantify the contribution of water depth, speed, load, and start-stop frequency to the acceleration of the propeller degradation, resulting in a low prediction accuracy and the inability to support maintenance decision optimization. More importantly, there is a technical gap between existing instantaneous diagnosis methods and predictive maintenance decisions, and the diagnosis results cannot be converted into specific maintenance time windows and working condition adjustment strategies. SUMMARY

[0005] The present application provides a propeller fault diagnosis and predictive maintenance method and system, which captures the high-order coupling relationship between propeller current, temperature, vibration, pressure and other parameters and its time evolution characteristics by constructing a multi-physical field coupling hypergraph time sequence encoder, and quantifies the influence of water depth, speed, load, and start-stop on degradation acceleration by combining a working condition sensitive degradation rate model, solving the problem that existing technologies cannot establish a fault evolution trajectory and residual life prediction.

[0006] In a first aspect, the present application provides a propeller fault diagnosis and predictive maintenance method, which comprises: Step S1: Collecting current signals, temperature signals, vibration signals and pressure signals of the propeller, extracting current harmonic distortion rate, temperature rise rate, vibration peak factor, vibration kurtosis factor and pressure response time, and constructing a multi-dimensional time sequence dataset; Step S2: Mapping the feature parameters in the multi-dimensional time sequence dataset as nodes, establishing electromechanical coupling hyperedges, thermal flow coupling hyperedges, mechanical degradation hyperedges and seal failure hyperedges, and constructing a hyperedge time sequence correlation matrix; Step S3: Inputting the hyperedge time sequence correlation matrix into the hypergraph convolution layer and time sequence attention layer of the multi-physical field coupling hypergraph time sequence encoder for forward propagation calculation, and outputting a degradation state feature vector; Step S4: Mapping the degradation state feature vector to a health index, performing time sequence differentiation on the health index to obtain a degradation rate reference function, calculating a working condition acceleration factor according to water depth, speed, load torque and start-stop frequency, and obtaining a degradation rate function; Step S5: integrating the reciprocal of the degradation rate function with respect to time, and terminating the integration when the health index reaches the failure threshold, to obtain a remaining service life prediction value.

[0007] In a second aspect, the application provides a propeller fault diagnosis and predictive maintenance system, comprising: A construction module is configured to collect current signals, temperature signals, vibration signals and pressure signals of the propeller, extract current harmonic distortion rate, temperature rise rate, vibration peak factor, vibration kurtosis factor and pressure response time, and construct a multi-dimensional time series dataset. A mapping module is configured to map feature parameters in the multi-dimensional time series dataset as nodes, establish electro-mechanical coupling hyper-edges, thermal flow coupling hyper-edges, mechanical degradation hyper-edges and sealing failure hyper-edges, and construct a hyper-edge time series correlation matrix. An input module is configured to input the hyper-edge time series correlation matrix into a hypergraph convolution layer and a time series attention layer of a multi-physical field coupling hypergraph time series encoder for forward propagation calculation, and output a degradation state feature vector. An output module is configured to map the degradation state feature vector as a health index, perform time series differentiation on the health index to obtain a degradation rate reference function, calculate a working condition acceleration factor according to water depth, speed, load torque and start-stop times, and obtain a degradation rate function. An operation module is configured to integrate the reciprocal of the degradation rate function with respect to time, and terminate the integration when the health index reaches the failure threshold, to obtain a remaining service life prediction value.

[0008] In a third aspect, a propeller fault diagnosis and predictive maintenance device is provided, comprising a memory and at least one processor, the memory storing instructions; the at least one processor calling the instructions in the memory to make the propeller fault diagnosis and predictive maintenance device execute the propeller fault diagnosis and predictive maintenance method described above.

[0009] In a fourth aspect, a computer readable storage medium is provided, the computer readable storage medium storing instructions, when executed on a computer, causing the computer to execute the propeller fault diagnosis and predictive maintenance method described above.

[0010] The technical scheme provided in the application overcomes the limitation of the prior art of collecting only single type sensor data or instantaneous state parameters, and realizes comprehensive characterization of the multi-physical field degradation characteristics of the propeller. The current harmonic distortion rate reflects the electromagnetic state of the magnetic coupling transmission and the motor winding, the temperature rise rate reflects the dynamic change of the motor thermal management capability, the vibration peak factor and the kurtosis factor capture the bearing impact characteristics and the wear distribution characteristics respectively, and the pressure response time reflects the dynamic compensation capability of the oil-filled sealing system. These parameters together constitute a feature system covering the mechanical, electrical, thermal and flow physical domains. The feature parameters in the multi-dimensional time series data set are mapped to nodes and the mechanical-electrical coupling hyperedge, the thermal-flow coupling hyperedge, the mechanical degradation hyperedge and the sealing failure hyperedge are established. The high-order relationship structure of the hyperedge connects multiple nodes, breaks through the limitation of the traditional method of establishing only a two-order relationship between pairs of nodes, and can explicitly characterize the multi-physical field coupling mechanisms such as current distortion causing motor heating, temperature rise affecting oil-filled pressure, and vibration peak and kurtosis co-evolution. When constructing the hyperedge time series correlation matrix, the time dimension is introduced and the correlation matrix elements are multiplied by an exponential decay time weight function, so that the weight of the historical data closer to the current time is higher, solving the problem that the existing static graph network cannot capture the time-varying characteristics of the topological relationship, and laying a data structure foundation for subsequent time series evolution modeling. The hyperedge time series correlation matrix is input into the supergraph convolution layer and the time series attention layer of the multi-physical field coupling supergraph time series encoder for forward propagation calculation to output a degradation state feature vector. The supergraph convolution layer realizes high-order propagation and aggregation of node features on the supergraph topological structure through multiple matrix multiplication operations of node degree normalization and hyperedge degree normalization, captures the spatial relationship of current, temperature, vibration and pressure parameters influencing each other through physical coupling mechanisms, and the time series attention layer calculates the key matrix, value matrix and executes attention weighted aggregation to automatically identify the most critical historical moment for the current degradation state judgment within the time window and give it a high weight, suppress the interference of unimportant moments, and solve the problem that the existing method equally processes all historical data, resulting in the key degradation information being submerged. The degradation state feature vector simultaneously fuses the multi-physical field spatial coupling information and the time series evolution trend information, and provides a high-quality feature representation for accurately predicting the fault evolution trajectory.

[0011] The degradation state feature vector is mapped to a health index, and a health index is time-differentiated to obtain a degradation rate benchmark function, a mapping relationship from an abstract feature space to an interpretable health metric is established, the health index quantifies the current degradation degree of the propeller in the interval of 0 to 1, and the degradation rate benchmark function reflects the natural degradation speed, thereby providing a benchmark reference for residual life prediction. When the working condition acceleration factor is calculated according to the water depth, the speed, the load torque and the start-stop times, the water depth square term is used to reflect the nonlinear accelerated aging of the sealing system under high pressure in deep water, the speed linear term is used to reflect the linear promotion of the fluid dynamic pressure on bearing wear, the load torque 1.5 power term is used to reflect the super-linear degradation of the magnetic coupling transmission under heavy load, and the logarithmic term of the start-stop times is used to reflect the marginal decreasing effect of the cumulative damage of start-stop impact. This working condition acceleration factor design based on physical driving breaks through the limitations of the existing methods assuming constant degradation rate or simple linear weighting, and can accurately quantify the differentiated contribution of different working condition factors to the degradation of the propeller. The degradation rate function is obtained by multiplying the degradation rate benchmark function and the working condition acceleration factor, the degradation rate is dynamically adjusted with the working condition, and the problem that the existing technology cannot associate the actual operating environment with the degradation speed is solved. The reciprocal of the degradation rate function is integrated with respect to time, and the integral is terminated when the health index reaches the failure threshold to obtain the residual service life prediction value. The trapezoidal rule numerical integration is used to ensure the calculation accuracy and efficiency, the integral termination condition is associated with the failure threshold to realize the direct mapping from the degradation trajectory to the maintenance time, the fundamental defect that the existing method can only diagnose the current state but cannot predict the future failure time is solved, the residual service life prediction value comprehensively considers the multi-physical field coupling degradation trend and the working condition modulation effect, provides an accurate time window reference for predictive maintenance decision-making, realizes the technical leap from passive fault diagnosis to active life prediction, and significantly reduces the risk of propeller sudden failure and optimizes the maintenance resource allocation. BRIEF DESCRIPTION OF DRAWINGS

[0012] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creating labor based on these drawings.

[0013] Figure 1 FIG. 1 is a schematic diagram of one embodiment of the propeller fault diagnosis and predictive maintenance method in the embodiments of the present application; Figure 2 FIG. 2 is a schematic diagram of one embodiment of the propeller fault diagnosis and predictive maintenance system in the embodiments of the present application; Figure 3 FIG. 3 is a structural schematic block diagram of the propeller fault diagnosis and predictive maintenance device in the embodiments of the present application. DETAILED DESCRIPTION

[0014] The embodiments of the present application provide a propeller fault diagnosis and predictive maintenance method and system. The terms "first", "second", "third", "fourth" and the like (if any) in the specification and claims of the present application and the above drawings are used to distinguish similar objects, and do not have to be used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the term "comprising" or "having" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0015] For ease of understanding, the specific process of the embodiments of the present application is described below. Please refer to Figure 1 One embodiment of the propeller fault diagnosis and predictive maintenance method in the embodiments of the present application includes: Step S1: Collecting current signals, temperature signals, vibration signals and pressure signals of the propeller, extracting current harmonic distortion rate, temperature rise rate, vibration peak value factor, vibration kurtosis factor and pressure response time, and constructing a multi-dimensional time series data set; Step S2: Mapping the feature parameters in the multi-dimensional time series data set as nodes, establishing electromechanical coupling super-edge, heat flow coupling super-edge, mechanical degradation super-edge and sealing failure super-edge, and constructing a super-edge time series correlation matrix; Step S3: Inputting the super-edge time series correlation matrix into the supergraph convolution layer and time series attention layer of the multi-physical field coupling supergraph time series encoder for forward propagation calculation, and outputting a degradation state feature vector; Step S4: Mapping the degradation state feature vector to a health index, performing time series differentiation on the health index to obtain a degradation rate reference function, calculating a working condition acceleration factor according to water depth, speed, load torque and start-stop times, and obtaining a degradation rate function; Step S5: Integrating the reciprocal of the degradation rate function with respect to time, terminating the integration when the health index reaches the failure threshold, and obtaining a remaining useful life prediction value.

[0016] It can be understood that the execution subject of the present application can be a propeller fault diagnosis and predictive maintenance system, and can also be a terminal or a server, which is not limited here. The embodiments of the present application take the server as the execution subject for example.

[0017] Specifically, the running data is collected by the built-in and external sensors of the propeller, the three-phase current waveform is obtained by the current sensor at a sampling frequency of 1000 Hz, the three-phase static coordinate system is converted into a two-phase rotating coordinate system by Park transformation to obtain the d-axis and q-axis current components, the three-phase system is converted into a two-phase orthogonal system by Clarke transformation, the sum of the square of the ratio of the amplitude of each harmonic to the amplitude of the fundamental wave in the d-axis and q-axis current components is calculated and then the square root is taken to obtain the total harmonic distortion rate as the current harmonic distortion rate, which reflects the motor winding state and magnetic coupling transmission efficiency. After the temperature sensor collects the motor winding temperature, the difference between the current temperature and the temperature at the previous time is divided by the time interval to obtain the temperature rise rate in a 10-second window. After the radial vibration data of the magnetic coupling transmission mechanism is collected by the vibration accelerometer, the time domain signal is converted into the frequency domain signal by fast Fourier transform to extract the vibration amplitude at the characteristic frequency, the ratio of the maximum value to the root mean square value of the amplitude sequence is calculated to obtain the vibration peak factor, and the ratio of the fourth central moment to the square of the second central moment is calculated to obtain the vibration kurtosis factor reflecting the bearing wear degree. The pressure sensor synchronously collects the internal pressure of the oil-filled sealed cabin and the external water pressure, and records the time interval from the start of the change to the stable value of the pressure difference after the differential operation of the two as the pressure response time reflecting the compensation ability of the sealing system. These parameters are arranged in time sequence and the water depth, speed, load torque and cumulative start-stop number at each time are recorded to form a multi-dimensional time series data set.

[0018] The five characteristic parameters of the multi-dimensional time series data set are mapped as nodes in the graph structure, the current harmonic distortion rate is mapped as the first node, the temperature rise rate is mapped as the second node, the vibration peak factor is mapped as the third node, the vibration kurtosis factor is mapped as the fourth node, and the pressure response time is mapped as the fifth node. A super-edge is established to connect multiple nodes to represent the multi-physical field coupling relationship, an electromechanical coupling super-edge connects the first node and the second node to represent the coupling process of current distortion causing motor heating, a heat flow coupling super-edge connects the second node and the fifth node to represent the coupling mechanism of temperature change affecting oil-filled pressure compensation, a mechanical degradation super-edge connects the third node and the fourth node to represent the vibration characteristic co-evolution caused by bearing wear, and a sealing failure super-edge also connects the second node and the fifth node to represent the joint action of temperature and pressure on sealing performance. When constructing the correlation matrix, if the node belongs to a super-edge, the corresponding element is 1, otherwise it is 0. After introducing the time dimension, each element of the correlation matrix is multiplied by an exponentially decaying time weight function, so that the historical data closer to the current time has a higher weight to obtain a super-edge time correlation matrix. This matrix contains both the spatial coupling topology between nodes and the time evolution information.

[0019] The super-edge time sequence correlation matrix is input into a multi-physical field coupling supergraph time sequence encoder for feature extraction. The encoder includes three modules, i.e., a supergraph convolution layer, a time sequence attention layer, and a feature fusion layer. The supergraph convolution layer first counts the total number of superedges connected to each node in the correlation matrix to form the diagonal elements of the node degree matrix, counts the total number of nodes connected by each superedge to form the diagonal elements of the superedge degree matrix, performs a negative two power operation on the node degree matrix to obtain a node degree normalization matrix, and performs a negative one power operation on the superedge degree matrix to obtain a superedge degree normalization matrix. The node degree normalization matrix, the superedge time sequence correlation matrix, the superedge weight matrix, the superedge degree normalization matrix, the transpose of the superedge time sequence correlation matrix, the node degree normalization matrix, the initial node feature matrix, and the trainable parameter matrix are multiplied in sequence to complete the supergraph convolution operation. The operation result is activated by a ReLU function to obtain a convolution feature matrix that captures the high-order spatial coupling relationship between nodes. The time sequence attention layer multiplies the convolution feature matrix by a query weight matrix, a key weight matrix, and a value weight matrix to obtain a query matrix, a key matrix, and a value matrix. The matrix product of the query matrix and the transpose of the key matrix is divided by the square root of the feature dimension, and then normalized by a softmax function to obtain an attention weight matrix. The weight matrix is multiplied by the value matrix to complete weighted aggregation to obtain an attention feature matrix that captures the correlation of features at different time points within a time window. The feature fusion layer concatenates the convolution feature matrix and the attention feature matrix along the feature dimension, and then inputs the concatenated matrix into a fully connected layer to reduce the dimension by matrix multiplication and bias addition to output a 128-dimensional degenerate state feature vector. The vector contains both multi-physical field spatial coupling information and time sequence evolution trend information.

[0020] The degradation state feature vector is input into a mapping layer, which includes a linear transformation to reduce the 128-dimensional vector to a 1-dimensional scalar, which is non-linearly mapped by a sigmoid function to the interval [0, 1] to obtain a health indicator, with a value of 1 indicating a completely healthy state and a value of 0 indicating a completely failed state. The time series of the health indicator is numerically differentiated to calculate the degradation rate reference function, which reflects the natural degradation speed of the thruster under the current working condition. The synchronous recorded working condition parameters are extracted from the multi-dimensional time series data set, and the square term of the ratio of water depth to the maximum working water depth of 6000 meters, the linear term of the ratio of speed to the maximum speed, the 1.5 power term of the ratio of load torque to the rated torque, and the logarithmic term of the ratio of the number of starts and stops to the threshold value of 1000 are calculated. The four terms are multiplied by weight coefficients 0.3, 0.2, 0.4, and 0.1, respectively, summed, and then added to 1 to obtain a working condition acceleration factor. The square term of the water depth reflects the non-linear accelerated aging of the sealing system under high pressure in deep water, the linear term of the speed reflects the promoting effect of the hydrodynamic pressure on bearing wear, the 1.5 power term of the load torque reflects the accelerated degradation of the magnetic coupling transmission under heavy load, and the logarithmic term of the number of starts and stops reflects the cumulative damage of start-stop impact. The degradation rate reference function is multiplied by the working condition acceleration factor to obtain the degradation rate function, which dynamically adjusts the real degradation speed of the thruster under the actual running environment.

[0021] The health index failure threshold is set to 0.15, and when the health index decreases to this value, it is determined that the thruster needs maintenance. The degradation time function is obtained by taking the reciprocal of the degradation rate function from the current time, which represents the time required for the health index to decrease by a unit value. The trapezoidal rule is used to numerically integrate the degradation time function along the time axis, which divides the integral interval into multiple small intervals, each with a step size of 1 hour. In each small interval, the integral value is approximated by the trapezoidal area, which is the arithmetic mean of the function values at the two ends of the interval multiplied by the interval width. All small interval trapezoidal areas are added to obtain the integral value. During the integration process, the health index value is continuously monitored, and when the health index first decreases to the failure threshold of 0.15, the integration operation is terminated. At this time, the time difference between the integration termination time and the current time is the predicted value of the remaining useful life. For example, the current health index of a certain underwater thruster is 0.82, indicating that it is in good condition. The degradation state feature vector extracted by the hypergraph convolution layer and the time attention layer is mapped to obtain this health index value. The health index sequence differential calculation of the past 100 hours obtains a degradation rate reference function of 0.003 per hour. The current working conditions are water depth of 4500 meters, speed of 80% of the rated value, load torque of 90% of the rated value, and cumulative start-stop of 800 times. The water depth term is 0.3 times 4500 divided by 6000 squared, which is equal to 0.169. The speed term is 0.2 times 0.8, which is equal to 0.16. The load term is 0.4 times 0.9 to the power of 1.5, which is equal to 0.343. The start-stop term is 0.1 times the natural logarithm 1.8, which is equal to 0.059. The four terms are added to the 1 to obtain the working condition acceleration factor, which is 1.731. The degradation rate function is 0.003 times 1.731, which is equal to a decrease of 0.0052 per hour. Taking the reciprocal of this degradation rate function gives 1.923 hours for each 0.01 decrease in health index. Integrating from the current 0.82 to the failure threshold of 0.15 requires a decrease of 0.67 in health index. Using the trapezoidal rule with a step size of 1 hour, the remaining useful life is approximately 129 hours. This predicted value takes into account the multi-physical field coupling degradation trend of current harmonics, temperature rise rate, vibration characteristics, and pressure response, as well as the acceleration effect of actual working conditions on the degradation rate.

[0022] In a specific embodiment, step S1 comprises: The three-phase current waveform data of the thruster is collected by a current sensor at a sampling frequency of 1000 Hz. The d-axis current component and the q-axis current component are extracted by performing Park transformation and Clark transformation on the three-phase current waveform data. The total harmonic distortion of the d-axis current component and the q-axis current component is calculated to obtain the current harmonic distortion rate. The temperature data of the thruster motor winding is collected by a temperature sensor. The time series differential operation is performed on the temperature data within a 10-second time window to obtain the temperature rise rate. The radial vibration acceleration data of the magnetic coupling transmission mechanism of the propeller is collected by a vibration accelerometer, vibration amplitude at a characteristic frequency is extracted by performing fast Fourier transform on the vibration acceleration data, a vibration peak factor is obtained by calculating a ratio of a peak value of the vibration amplitude to a root mean square value, and a vibration kurtosis factor is obtained by calculating a ratio of a fourth moment of the vibration amplitude to a square of a second moment; The internal pressure signal and the external water pressure signal of the propeller oil-filled sealed cabin are collected by a pressure sensor, a differential operation is performed on the internal pressure signal and the external water pressure signal, a time interval from a change starting point to a stable ending point of a pressure difference value is calculated, and a pressure response time is obtained; The current harmonic distortion rate, the temperature rise rate, the vibration peak factor, the vibration kurtosis factor and the pressure response time are organized according to a time sequence, the water depth, the speed, the load torque and the cumulative start-stop number at the corresponding time are recorded synchronously, and a multi-dimensional time sequence data set is constructed.

[0023] Specifically, the current sensor continuously acquires the instantaneous values of the three-phase current of the propeller motor at a sampling frequency of 1000 times per second, obtaining a three-phase current waveform data sequence. The Park transformation converts the current components in the three-phase stationary coordinate system to the two-phase rotating coordinate system, specifically, the A-phase, B-phase, and C-phase three-phase currents are mapped to the d-axis current component and the q-axis current component through the coordinate transformation matrix, where the d-axis is consistent with the magnetic field direction and the q-axis is perpendicular to the magnetic field direction. The Clarke transformation converts the three-phase coordinate system to the two-phase orthogonal stationary coordinate system, and through two transformations, the three-phase current signal is simplified to two-phase components for subsequent analysis. The d-axis current component and the q-axis current component are respectively subjected to Fourier decomposition to extract the harmonic components, and the total harmonic distortion rate is calculated by taking the square root of the sum of the squares of all harmonic amplitudes excluding the fundamental wave divided by the square of the fundamental wave amplitude. The larger the parameter value, the more unstable the motor operates and the lower the magnetic coupling transmission efficiency. The temperature sensor continuously acquires the surface temperature of the motor winding, records the temperature value at the start of the window and the temperature value at the end of the window in a 10-second time window, and the temperature change is obtained by subtracting the two. The temperature rise rate is obtained by dividing the 10-second time interval, which reflects the motor heating speed and heat dissipation capacity. The vibration accelerometer acquires the radial direction vibration acceleration time domain signal of the magnetic coupling transmission mechanism shell, and the fast Fourier transform converts the time domain signal to the frequency domain signal to obtain the vibration amplitude corresponding to different frequencies. The vibration peak factor is calculated by dividing the maximum amplitude in the amplitude sequence by the root mean square value of the amplitude sequence, and the root mean square value is obtained by dividing the sum of the squares of all amplitudes by the number of amplitudes. The larger the peak factor value, the more prominent the impact vibration. The vibration kurtosis factor is calculated by dividing the fourth-order central moment of the amplitude sequence by the square of the second-order central moment, and the fourth-order central moment is the average of the fourth power of each amplitude minus the average value. The second-order central moment is the average of the square of each amplitude minus the average value, i.e. the variance. The larger the kurtosis factor value, the sharper the amplitude distribution and the more serious the bearing wear. The pressure sensor simultaneously acquires the internal pressure of the oil-filled sealed cabin and the external water pressure of the propeller, and the pressure difference sequence is obtained by subtracting the two pressure signals. When the external water pressure changes, the internal oil-filled system needs to automatically compensate the pressure to maintain balance. The time interval from the start of the pressure difference deviating from the equilibrium point to the re-achieve of the equilibrium state is recorded as the pressure response time. The longer the response time, the worse the performance of the oil-filled pressure compensation system and the higher the risk of seal failure.

[0024] The five characteristic parameters are arranged in time sequence according to the collection time sequence, and the current harmonic distortion rate value, the temperature rise rate value, the vibration peak factor value, the vibration kurtosis factor value and the pressure response time value are recorded at each time. At the same time, the working condition parameters of the propeller are recorded at each time. The water depth is measured by a depth gauge to obtain the current underwater depth of the propeller. The speed is calculated by the propeller speed sensor and the propeller parameter to obtain the sailing speed. The load torque is calculated by the motor current and voltage to obtain the motor output torque. The cumulative start-stop number is recorded by a counter to record the total number of start-stop cycles of the propeller from the start of use to the current time. The five characteristic parameter sequences and the four working condition parameter sequences are aligned by time to form a multi-dimensional time sequence data set. Each row of the data set corresponds to a sampling time, each column corresponds to a parameter dimension, and nine columns of data form a state-condition associated data set.

[0025] In a specific embodiment, step S2 comprises: The current harmonic distortion rate, the temperature rise rate, the vibration peak factor, the vibration kurtosis factor and the pressure response time in the multi-dimensional time sequence data set are respectively mapped to a first node, a second node, a third node, a fourth node and a fifth node, and a node set is constructed. An electromechanical coupling hyperedge connecting the first node and the second node is established, a heat flow coupling hyperedge connecting the second node and the fifth node is established, a mechanical degradation hyperedge connecting the third node and the fourth node is established, and a sealing failure hyperedge connecting the second node and the fifth node is established, and a hyperedge set is constructed. According to the node set and the hyperedge set, an association matrix is constructed, wherein if the node belongs to the hyperedge, the corresponding element of the association matrix is valued as 1, otherwise, the value is 0. The time dimension of the association matrix is expanded, a preset time window length is set, each element of the association matrix is multiplied by a time weight function, the time weight function is an exponential decay function, and a hyperedge time sequence association matrix is obtained.

[0026] Specifically, the multi-dimensional time series dataset contains time series of five feature parameters, the current harmonic distortion rate sequence is mapped to the first node representing the electromagnetic state of the motor, the temperature rise rate sequence is mapped to the second node representing the thermal state of the motor, the vibration peak factor sequence is mapped to the third node representing the vibration impact characteristics, the vibration kurtosis factor sequence is mapped to the fourth node representing the vibration distribution characteristics, and the pressure response time sequence is mapped to the fifth node representing the sealing system state. The five nodes form a node set. The hyperedge is a high-order relationship structure connecting multiple nodes. The electromechanical coupling hyperedge connects the first node and the second node to represent the physical coupling relationship that current distortion causes motor heating. The heat flow coupling hyperedge connects the second node and the fifth node to represent the coupling relationship that temperature rise leads to oil expansion affecting pressure compensation. The mechanical degradation hyperedge connects the third node and the fourth node to represent the coupling relationship that bearing wear simultaneously affects vibration peak and kurtosis. The sealing failure hyperedge connects the second node and the fifth node to represent the coupling relationship that temperature and pressure jointly act on the aging of sealing materials. The four hyperedges form a hyperedge set.

[0027] The incidence matrix is a two-dimensional matrix describing the connection relationship between nodes and hyperedges. The number of matrix rows is equal to the number of nodes, a total of five rows, and the number of matrix columns is equal to the number of hyperedges, a total of four columns. Traverse each combination of nodes and hyperedges, and determine whether the node is connected by the hyperedge. The first node belongs to the electromechanical coupling hyperedge, and the first element of the first row and the first column of the incidence matrix is 1. The first node does not belong to the heat flow coupling hyperedge, and the first element of the first row and the second column of the incidence matrix is 0. All elements of the matrix are filled in turn to obtain the complete incidence matrix. The first node belongs to the electromechanical coupling hyperedge, and the corresponding element is 1, and does not belong to the remaining three hyperedges, and the corresponding element is 0. The second node belongs to the electromechanical coupling hyperedge, the heat flow coupling hyperedge, and the sealing failure hyperedge, and the corresponding element is 1, and does not belong to the mechanical degradation hyperedge, and the corresponding element is 0. The third node belongs to the mechanical degradation hyperedge, and the corresponding element is 1, and does not belong to the remaining three hyperedges, and the corresponding element is 0. The fourth node belongs to the mechanical degradation hyperedge, and the corresponding element is 1, and does not belong to the remaining three hyperedges, and the corresponding element is 0. The fifth node belongs to the heat flow coupling hyperedge and the sealing failure hyperedge, and the corresponding element is 1, and does not belong to the remaining two hyperedges, and the corresponding element is 0. The sparse incidence matrix reflects the topological connection mode of nodes and hyperedges.

[0028] The time dimension is introduced to expand the two-dimensional correlation matrix into a three-dimensional hyperedge time-series correlation matrix, and the third dimension is the length of the time axis equal to the length of the preset time window. The time weight function adopts an exponential decay function form, and the function value decreases with the increase of the time interval. The closer the historical data to the current time, the higher the weight, and the farther the historical data from the current time, the lower the weight. Multiply each element of the correlation matrix by the time weight function value of the corresponding time. The correlation matrix element of the current time is multiplied by the weight value 1 to keep the original value. The correlation matrix element of the previous time is multiplied by a weight value less than 1 to reduce the value. The correlation matrix element of the earlier time is multiplied by a smaller weight value to further reduce the value. Traverse all time points in the time window, multiply all elements of the correlation matrix at each time point by the time weight function value corresponding to the time point, and stack the weighted correlation matrix in time order to form a three-dimensional hyperedge time-series correlation matrix. The first dimension of the matrix is the node dimension, the second dimension is the hyperedge dimension, and the third dimension is the time dimension. The matrix element value reflects the connection relationship between nodes and hyperedges and the weight relationship of time proximity.

[0029] In a specific embodiment, step S3 comprises: The hyperedge time-series correlation matrix is input into the hypergraph convolution layer of the multi-physical field coupled hypergraph time-series encoder to calculate the node degree matrix and the hyperedge degree matrix, and the hypergraph convolution operation is performed on the hyperedge time-series correlation matrix to obtain the convolution feature matrix. The convolution feature matrix is input into the time-series attention layer of the multi-physical field coupled hypergraph time-series encoder to calculate the query matrix, key matrix and value matrix at different time points in the time window, calculate the attention weight according to the query matrix and key matrix, and perform weighted aggregation on the value matrix to obtain the attention feature matrix. The convolution feature matrix and the attention feature matrix are spliced and input into the feature fusion layer of the multi-physical field coupled hypergraph time-series encoder for dimension reduction processing to output the degradation state feature vector.

[0030] Specifically, after the superedge time correlation matrix is input into the supergraph convolution layer of the multi-physical field coupled supergraph time encoder, first, the node degree matrix and the superedge degree matrix reflecting the topological properties of the graph structure are calculated. The node degree matrix is a diagonal matrix, and each element on the diagonal line represents the total number of superedges connected to the corresponding node. By traversing each row of the superedge time correlation matrix, the sum of all non-zero elements in the row is obtained to fill in the degree value of the node on the diagonal line of the node degree matrix. The superedge degree matrix is also a diagonal matrix, and each element on the diagonal line represents the total number of nodes connected to the corresponding superedge. By traversing each column of the superedge time correlation matrix, the sum of all non-zero elements in the column is obtained to fill in the degree value of the superedge on the diagonal line of the superedge degree matrix. The negative square root of two of each diagonal element of the node degree matrix is taken to obtain the node degree normalization matrix, and the negative power of each diagonal element of the superedge degree matrix is taken to obtain the superedge degree normalization matrix. The supergraph convolution operation multiplies the node degree normalization matrix on the left of the superedge time correlation matrix, and then multiplies the result on the right of the superedge weight matrix, which is a diagonal matrix with diagonal elements representing the importance weight of each superedge. Then, the result is multiplied on the right of the superedge degree normalization matrix, and then multiplied on the right of the transpose matrix of the superedge time correlation matrix, which exchanges rows and columns. Then, the result is multiplied on the right of the node degree normalization matrix, and then multiplied on the right of the initial node feature matrix, which has each row corresponding to a node and each column corresponding to a feature dimension. Finally, the result is multiplied on the right of the trainable parameter matrix to complete the linear transformation, and the parameter matrix is obtained through training and learning. The results of all matrix multiplications are input into the ReLU activation function for nonlinear transformation, and the ReLU function sets negative values to zero and preserves positive values. The output convolution feature matrix contains the feature representation of the nodes after propagation through the superedges.

[0031] After the convolution feature matrix is input into the time attention layer of the multi-physical field coupled supergraph time encoder, it is multiplied by three trainable weight matrices respectively to generate query matrix, key matrix and value matrix. The query matrix is obtained by multiplying the convolution feature matrix on the right of the query weight matrix, the key matrix is obtained by multiplying the convolution feature matrix on the right of the key weight matrix, and the value matrix is obtained by multiplying the convolution feature matrix on the right of the value matrix. The three matrices have the same dimensions but different contents, representing query information, key information and value information in the attention mechanism respectively. The matrix product of the query matrix and the transpose of the key matrix is calculated, and each element of the product matrix represents the similarity of the features at two time points. Each element of the product matrix is divided by the square root of the feature dimension to avoid large numerical values. The scaled matrix is input into the softmax function for normalization. The softmax function performs exponential operation on each row of the matrix and divides by the sum of the exponents of all elements in the row to make the sum of the elements in each row equal to 1. The normalized matrix is the attention weight matrix, and each element of the matrix represents the attention allocation weight of a time point to another time point. The attention weight matrix is multiplied by the value matrix to complete the weighted aggregation. Each row of the weight matrix is multiplied by the value matrix to obtain the aggregated feature vector. The aggregation results of all time points form the attention feature matrix, which highlights the features of important time points and suppresses the features of unimportant time points within the time window.

[0032] The convolution feature matrix and the attention feature matrix are spliced along the feature dimension. The splicing combines the two matrices in the column direction. If the convolution feature matrix has 64 features per row and the attention feature matrix has 64 features per row, the spliced matrix has 128 features per row. The spliced matrix contains spatial coupling information and time sequence correlation information. The spliced matrix is input into the feature fusion layer of the multi-physical field coupling hypergraph time sequence encoder. The fusion layer includes a full connection layer for dimension reduction processing. The full connection layer completes linear transformation by right multiplication of the spliced matrix by a dimension reduction weight matrix and addition of a bias vector. The dimension reduction weight matrix has fewer rows than columns to realize feature dimension compression. The linear transformation outputs a degradation state feature vector. The feature vector has a fixed dimension of 128. Each element of the vector is an abstract representation extracted from the original multi-dimensional features. The feature vector contains high-order coupling relationships between multi-physical field parameters such as current, temperature, vibration, and pressure, and evolution trend information of the parameters on the time axis.

[0033] In a specific embodiment, the hyperedge time sequence correlation matrix is input into the hypergraph convolution layer of the multi-physical field coupling hypergraph time sequence encoder. A node degree matrix and a hyperedge degree matrix are calculated. The hypergraph convolution operation is performed on the hyperedge time sequence correlation matrix to obtain a convolution feature matrix, including: inputting the hyperedge time sequence correlation matrix into the hypergraph convolution layer of the multi-physical field coupling hypergraph time sequence encoder; counting the number of hyperedges connected to each node in the hyperedge time sequence correlation matrix to construct the diagonal elements of the node degree matrix; counting the number of nodes connected to each hyperedge in the hyperedge time sequence correlation matrix to construct the diagonal elements of the hyperedge degree matrix; performing a negative two power operation on the node degree matrix to obtain a node degree normalization matrix, and performing a negative one power operation on the hyperedge degree matrix to obtain a hyperedge degree normalization matrix; performing matrix multiplication operations on the node degree normalization matrix, the hyperedge time sequence correlation matrix, the hyperedge weight matrix, the hyperedge degree normalization matrix, the transpose matrix of the hyperedge time sequence correlation matrix, the node degree normalization matrix, the initial node feature matrix, and the trainable parameter matrix in sequence. The ReLU activation function is applied to the operation result to obtain the convolution feature matrix.

[0034] Specifically, the superedge time sequence correlation matrix starts the preparation stage of supergraph convolution operation after the supergraph convolution layer of the multi-physical field coupling supergraph time sequence encoder. Each row of the superedge time sequence correlation matrix is traversed, and the number of elements with a value of 1 in the row, i.e. the number of superedges connected to the node, is counted. The first node corresponds to a row with only the mechanical-electric coupling superedge column element being 1 and the rest of the columns being 0, so the degree value of the first node is 1. The second node corresponds to a row with the mechanical-electric coupling superedge column, the heat flow coupling superedge column and the sealing failure superedge column elements being 1, so the degree value of the second node is 3. The third node corresponds to a row with only the mechanical degradation superedge column element being 1, so the degree value is 1. The fourth node corresponds to a row with only the mechanical degradation superedge column element being 1, so the degree value is 1. The fifth node corresponds to a row with the heat flow coupling superedge column and the sealing failure superedge column elements being 1, so the degree value is 2. The five degree values are sequentially filled into the diagonal line of the node degree matrix to form a diagonal matrix, and the rest of the elements are 0. Each column of the superedge time sequence correlation matrix is traversed, and the number of elements with a value of 1 in the column, i.e. the number of nodes connected to the superedge, is counted. The mechanical-electric coupling superedge column has the first node row and the second node row elements being 1, so the degree value is 2. The heat flow coupling superedge column has the second node row and the fifth node row elements being 1, so the degree value is 2. The mechanical degradation superedge column has the third node row and the fourth node row elements being 1, so the degree value is 2. The sealing failure superedge column has the second node row and the fifth node row elements being 1, so the degree value is 2. The four degree values are sequentially filled into the diagonal line of the superedge degree matrix to form a diagonal matrix, and the rest of the elements are 0.

[0035] The negative two power operation is performed on each diagonal element of the node degree matrix. The negative two power of the first node degree value 1 is equal to 1, the negative two power of the second node degree value 3 is equal to the square root of 3, the negative two power of the third node degree value 1 is equal to 1, the negative two power of the fourth node degree value 1 is equal to 1, and the negative two power of the fifth node degree value 2 is equal to the square root of 2. The operation results are filled into the diagonal line of the node degree normalization matrix to obtain the normalization matrix. The negative power operation, i.e. taking the reciprocal, is performed on each diagonal element of the superedge degree matrix. The negative power of the mechanical-electric coupling superedge degree value 2 is equal to one half, the negative power of the heat flow coupling superedge degree value 2 is equal to one half, the negative power of the mechanical degradation superedge degree value 2 is equal to one half, and the negative power of the sealing failure superedge degree value 2 is equal to one half. The operation results are filled into the diagonal line of the superedge degree normalization matrix to obtain the normalization matrix. The normalization processing reduces the weights of nodes and superedges with large degree values and increases the weights of nodes and superedges with small degree values, balancing the contributions of nodes and superedges in different topologies.

[0036] The sequential matrix multiplication operation is performed. First, the node degree normalization matrix is multiplied with the super-edge time correlation matrix on the left. The matrix multiplication is calculated according to the row-column product rule. The node degree normalization matrix is a diagonal matrix multiplied on the left, which is equivalent to scaling the corresponding row of the correlation matrix with the diagonal elements. The first row is multiplied by 1 and remains unchanged. The second row is multiplied by the square root of 3 and is reduced by one. The first intermediate result matrix is obtained. The first intermediate result matrix is multiplied with the super-edge weight matrix on the right. The super-edge weight matrix is also a diagonal matrix, and the diagonal elements represent the importance weight of each super-edge. The right multiplication is equivalent to scaling the corresponding column of the intermediate result matrix with the diagonal elements. Each column is adjusted according to the super-edge weight to obtain the second intermediate result matrix. The second intermediate result matrix is multiplied with the super-edge degree normalization matrix on the right. The diagonal matrix is multiplied on the right to scale the column elements. Each column is multiplied by the reciprocal of the corresponding super-edge degree to complete the super-edge side normalization to obtain the third intermediate result matrix. The third intermediate result matrix is multiplied with the transpose matrix of the super-edge time correlation matrix on the right. The transpose matrix exchanges the rows and columns of the original matrix. This step realizes the aggregation process of information from the super-edge back to the node. Each column of the third intermediate result matrix represents the information weighted by the super-edge. After multiplying the transpose matrix, each node aggregates information from different super-edges to obtain the fourth intermediate result matrix. The fourth intermediate result matrix is multiplied with the node degree normalization matrix on the right. The diagonal matrix is multiplied on the right to scale the column elements to complete the node side normalization to obtain the fifth intermediate result matrix. The fifth intermediate result matrix is multiplied with the initial node feature matrix on the right. The initial node feature matrix is composed of the original feature vector of each node, which is normalized by parameters such as current harmonic distortion rate and temperature rise rate. The matrix multiplication applies the topological propagation weight to the node feature to obtain the sixth intermediate result matrix. The sixth intermediate result matrix is multiplied with the trainable parameter matrix on the right. The trainable parameter matrix is learned from the training data through the back propagation algorithm. The matrix multiplication completes the linear transformation of the feature space to map to a new representation space to obtain the final matrix multiplication result.

[0037] The ReLU activation function is applied to each element of the final matrix multiplication result. The ReLU function is defined as zero when the input is less than zero, and the output is equal to the input when the input is greater than or equal to zero. All elements of the result matrix are traversed. If the element value is negative, the element is set to zero. If the element value is positive, the element remains unchanged. The ReLU activation introduces a nonlinear transformation capability to enable the network to learn complex nonlinear mapping relationships. After ReLU processing, the convolution feature matrix is obtained. Each row of the matrix corresponds to a node, and each column corresponds to a feature dimension. The matrix element value reflects the representation of the node in the feature dimension after topological propagation and nonlinear transformation of the supergraph. The convolution feature matrix integrates the node's own features and the features transmitted by the neighbor nodes through the super-edge, capturing high-order feature representations related by physical mechanisms such as electromechanical coupling and heat flow coupling between current temperature vibration pressure and other multi-physical field parameters.

[0038] In a specific embodiment, step S4 comprises: The degradation state feature vector is input into a mapping layer, and a health index is obtained through nonlinear mapping by a sigmoid function; A numerical differential operation is performed on the time series of the health index to calculate a ratio of a difference value of the health index at adjacent time instants to a time interval, and a degradation rate reference function is obtained. According to the water depth, the speed, the load torque, and the cumulative start-stop number recorded in the multi-dimensional time series data set, a square term of a ratio of the water depth to the maximum working water depth, a linear term of a ratio of the speed to the maximum speed, a 1.5 power term of a ratio of the load torque to the rated torque, and a logarithmic term of a ratio of the start-stop number to a threshold value are calculated, a working condition acceleration factor is obtained by weighted summation of the terms and adding 1. The degradation rate reference function is multiplied by the working condition acceleration factor to obtain a degradation rate function.

[0039] Specifically, the degradation state feature vector is a 128-dimensional vector containing multi-physical field coupling degradation information of the propeller. After being input into the mapping layer, the 128-dimensional vector is first projected into a 1-dimensional scalar through linear transformation. The linear transformation is completed by point multiplication of the vector and a weight vector and adding a bias value. The weight vector and the bias value are obtained through training and learning. The scalar value domain range is from negative infinity to positive infinity. The scalar is input into a sigmoid function for nonlinear mapping. The sigmoid function expression is that the output is equal to 1 divided by 1 plus the negative input power of the natural constant e. When the input is negative infinity, the output tends to 0. When the input is 0, the output is equal to 0.5. When the input is positive infinity, the output tends to 1. The sigmoid function maps any real number to the interval [0, 1] to obtain the health index. A health index value of 1 indicates a completely healthy state of the propeller, and a value of 0 indicates a completely failed state. Intermediate values indicate different degrees of degradation.

[0040] A numerical differential operation is performed on the time series of the health index to calculate a ratio of a difference value of the health index at adjacent time instants to a time interval, and a degradation rate reference function is obtained.

[0041] According to the working condition parameters recorded in the multi-dimensional time series data set, the acceleration effect of the working condition on the degradation rate is calculated, the water depth value corresponding to the current time is extracted, and the water depth ratio is obtained by dividing the maximum working water depth. The water depth ratio is squared to obtain the water depth square term. The maximum working water depth of the propeller is 6000 meters, and the water depth square term reflects the nonlinear accelerated aging effect of high pressure in deep water on the sealing system. The deeper the water depth, the greater the pressure, and the faster the aging of the sealing material. The square relationship indicates that the acceleration effect is nonlinearly enhanced with the depth. The speed value is extracted and divided by the maximum speed to obtain the speed ratio as the speed linear term. The speed linear term reflects the promoting effect of fluid dynamic pressure on bearing wear. The higher the speed, the faster the propeller speed, and the greater the load on the bearing, the faster the wear. The linear relationship indicates that the wear rate is proportional to the speed. The load torque value is extracted and divided by the rated torque to obtain the load ratio. The load ratio is raised to the power of 1.5 to obtain the load 1.5 power term. The load 1.5 power term reflects the accelerated degradation characteristics of the magnetic coupling transmission under heavy load. The greater the load torque, the greater the torque on the magnetic coupling mechanism, and the faster the transmission efficiency decreases. The 1.5 power indicates that the degradation rate increases superlinearly with the load. The cumulative start-stop number is extracted and divided by the threshold value of 1000 times to obtain the start-stop ratio. The start-stop ratio is added to 1 and the natural logarithm is taken to obtain the start-stop logarithmic term. The start-stop logarithmic term reflects the cumulative damage effect of start-stop impact. The more the start-stop number, the more frequent the start-stop impact on the motor and the magnetic coupling mechanism, and the more serious the cumulative damage. The logarithmic relationship indicates that the marginal damage effect decreases. The water depth square term is multiplied by the weight coefficient 0.3, the speed linear term is multiplied by the weight coefficient 0.2, the load 1.5 power term is multiplied by the weight coefficient 0.4, and the start-stop logarithmic term is multiplied by the weight coefficient 0.1. The weighted sum of the four terms plus 1 obtains the working condition acceleration factor. The plus 1 operation ensures that the minimum value of the acceleration factor is 1, indicating that there is no acceleration under the rated working condition. Greater than 1 indicates that the degradation is accelerated in harsh working conditions. The weight coefficient reflects the relative contribution of different working condition factors to the degradation of the propeller.

[0042] The degradation rate function is obtained by multiplying the degradation rate benchmark function and the working condition acceleration factor at each time point. The actual degradation rate is obtained by multiplying the natural degradation rate and the working condition modulation coefficient. The value of the degradation rate function at each time point is equal to the product of the benchmark degradation rate at the time point and the working condition acceleration factor at the time point. When the thruster is operated in harsh conditions such as deep water, high load, and frequent start-stop, the working condition acceleration factor is large, and the value of the degradation rate function is large, indicating that the degradation is fast. When the thruster is operated in good conditions such as shallow water, low load, and stable operation, the working condition acceleration factor is close to 1, and the value of the degradation rate function is close to the benchmark value, indicating that the degradation is slow. The health index is obtained by linear transformation of the degradation state feature vector of the thruster in a certain deep sea operation through the mapping layer and the sigmoid function mapping. The health index gradually decreases from a high value during continuous operation, reflecting the degradation of the thruster. The derivative of the time series of the health index shows that the health index at the current time is lower than the health index at the previous time, and the difference between the two is divided by the time interval to obtain a negative degradation rate benchmark function representing the decreasing trend of the health index. At the same time, the working condition parameters are extracted to find that the thruster is located in the deep water area with a water depth higher than the rated water depth. The water depth is divided by the maximum working water depth and then squared to obtain a water depth square term greater than the rated working condition. The speed is close to the maximum speed to obtain a larger speed linear term. The load torque exceeds the rated torque to obtain a larger load 1.5 power term. The cumulative start-stop frequency is high to obtain a certain value of the start-stop logarithmic term. The weighted sum of the four terms is added by 1 to obtain a working condition acceleration factor greater than 1, indicating that the current harsh working condition accelerates the degradation of the thruster. The degradation rate function obtained by multiplying the degradation rate benchmark function and the working condition acceleration factor has a larger absolute value, reflecting that the degradation speed of the thruster in the current deep water and high load working condition is significantly faster than that in the normal working condition.

[0043] In a specific embodiment, step S5 comprises: The failure threshold is set to 0.15, and when the health index decreases to the failure threshold, it is determined that the thruster needs maintenance; Starting from the current time, the degradation rate function is operated in reverse to obtain the degradation time function; The degradation time function is integrated along the time axis, and the trapezoidal rule is used for numerical integration with an integral step size of 1 hour; When the health index reaches the failure threshold, the integral operation is terminated, and the time difference between the integral termination time and the current time is taken as the remaining service life prediction value.

[0044] Specifically, the failure threshold is set to 0.15 as the critical value of the health index of the propeller requiring maintenance, which is determined based on historical failure data of the propeller, and when the health index falls below 0.15, the degradation level of key components of the propeller such as bearing wear, seal failure, and magnetic coupling transmission efficiency reduction reaches a level affecting safe operation, and maintenance and component replacement are required. The health index is continuously monitored from the current time, and the health index value is calculated at each sampling time, and when the health index at a certain time is first below 0.15, the maintenance determination is triggered.

[0045] The degradation rate function is operated in reverse from the current time to obtain the degradation time function, and each time value of the degradation rate function represents the speed of the health index at that time. The degradation rate is negative, indicating that the health index is decreasing, and the larger the absolute value of the degradation rate, the faster the decrease. The reverse operation takes the reciprocal of each time value of the degradation rate function, and the negative value of the degradation rate remains negative after the reciprocal operation, but the physical meaning is changed to the time required for the health index to decrease by a unit value. When the absolute value of the degradation rate is large, the absolute value of the reciprocal is small, indicating that the decrease is fast and the time required is short. When the absolute value of the degradation rate is small, the absolute value of the reciprocal is large, indicating that the decrease is slow and the time required is long. The degradation time function sequence is obtained by performing the reciprocal operation on each time value of the degradation rate function sequence, and the degradation time function reflects the time consumption rate of the propeller from the current health state to the failure state at different times.

[0046] The total time consumption is calculated by integrating the degradation time function along the time axis, and the integral operation calculates the area enclosed by the function curve and the time axis. When the degradation time function is negative, the integral value is negative, indicating the time accumulation in the direction of the health index decrease. Numerical integration is performed using the trapezoidal rule, which divides the integral interval into multiple small intervals, and the integral step is set to 1 hour, indicating that the width of each small interval is 1 hour. In each small interval, the trapezoidal area is used to approximate the integral value of the interval. The trapezoidal area calculation method is to take the degradation time function values at the start and end times of the interval, add them together and divide by 2 to obtain the average height of the trapezoid, and then multiply the average height by the interval width of 1 hour to obtain the trapezoidal area of the small interval. Starting from the current time and moving forward in the future direction, the first small interval is from the current time to the current time plus 1 hour, and the degradation time function values at the start and end times of the interval are calculated to obtain the average value, which is multiplied by 1 hour to obtain the first trapezoidal area. The second small interval is from the current time plus 1 hour to the current time plus 2 hours, and the second trapezoidal area is calculated, and each small interval is calculated and accumulated in turn.

[0047] In the integral process, the health index value corresponding to each time is calculated synchronously, the initial value of the health index is the health index at the current time, the health index decreases by the degradation amount corresponding to the interval after each interval is pushed, the degradation amount is equal to the degradation rate of the interval multiplied by the time interval of 1 hour, and the accumulated degradation amount continuously increases to cause the health index to continuously decrease. After each interval integration is completed, it is checked whether the health index at the time is lower than the failure threshold of 0.15, if the health index is still higher than 0.15, the next interval integration is continued, if the health index first decreases to 0.15 or below, the integration operation is terminated, and the termination time is the predicted failure time. The time difference between the termination time and the current time is calculated, the time difference is equal to the number of intervals that have completed integration multiplied by the integral step of 1 hour, and the time difference is taken as the remaining useful life prediction value to represent the remaining available time of the thruster from the current time to the time when maintenance is needed. The health index of a certain thruster is higher than the failure threshold, the degradation rate function is negative, reflecting that the health index continuously decreases, the reciprocal of the degradation rate function is also negative, the trapezoidal rule integration starts from the current time to calculate a trapezoidal area every 1 hour, the absolute value of the degradation time function value in the first interval is larger, indicating that the decrease is slow, the absolute value of the degradation time function value in the second interval is smaller due to the change of the working condition, indicating that the decrease is fast, the trapezoidal areas are accumulated and the health index is continuously decreased, and the integration is continuously performed until the health index decreases to 0.15 at a certain time to terminate the integration. The time difference between the time and the current time is the remaining useful life prediction value, reflecting the remaining available time window of the thruster under the current degradation trend and working condition.

[0048] The thruster fault diagnosis and predictive maintenance method in the embodiments of the present application is described above, and the thruster fault diagnosis and predictive maintenance system in the embodiments of the present application is described below. Please refer to Figure 2 One embodiment of the thruster fault diagnosis and predictive maintenance system in the embodiments of the present application includes: A construction module is configured to collect current signals, temperature signals, vibration signals and pressure signals of the thruster, extract current harmonic distortion rate, temperature rise rate, vibration peak value factor, vibration kurtosis factor and pressure response time, and construct a multi-dimensional time series data set. A mapping module is configured to map feature parameters in the multi-dimensional time series data set into nodes, establish electromechanical coupling super edges, thermal flow coupling super edges, mechanical degradation super edges and sealing failure super edges, and construct a super edge time series correlation matrix. An input module is configured to input the super edge time series correlation matrix into a supergraph convolution layer and a time series attention layer of a multi-physical field coupling supergraph time series encoder to perform forward propagation calculation, and output a degradation state feature vector. The output module is configured to map the degradation state feature vector into a health index, to obtain a degradation rate reference function by time series differentiation of the health index, to obtain a working condition acceleration factor by calculating water depth, speed, load torque and start-stop times, and to obtain a degradation rate function; The operation module is configured to perform integral operation on time with an inverse of the degradation rate function, to terminate the integral operation when the health index reaches a failure threshold, and to obtain a remaining service life prediction value.

[0049] The above Figure 2 The propeller fault diagnosis and predictive maintenance system in the embodiment of the present application is described in detail from the perspective of modular functional entities, and the propeller fault diagnosis and predictive maintenance device in the embodiment of the present application is described in detail from the perspective of hardware processing.

[0050] Referring to Figure 3 The embodiment of the present application also provides a propeller fault diagnosis and predictive maintenance device, which can be a server, and the internal structure of the propeller fault diagnosis and predictive maintenance device can be as shown in Figure 3 The propeller fault diagnosis and predictive maintenance device includes a processor, a memory, a display screen, an input device, a network interface and a database connected through a system bus. The processor of the computer is configured to provide computing and control capabilities. The memory of the propeller fault diagnosis and predictive maintenance device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The database of the propeller fault diagnosis and predictive maintenance device is configured to store corresponding data in the embodiment. The network interface of the propeller fault diagnosis and predictive maintenance device is configured to communicate with an external terminal through a network connection. The computer program is executed by the processor to implement the above method.

[0051] Those skilled in the art can understand, Figure 3 The structure shown in the embodiment of the present application is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the propeller fault diagnosis and predictive maintenance device to which the scheme of the present application is applied.

[0052] The present application also provides a computer readable storage medium, which can be a non-volatile computer readable storage medium or a volatile computer readable storage medium. The computer readable storage medium stores instructions, and when the instructions are run on a computer, the computer executes the steps of the propeller fault diagnosis and predictive maintenance method.

[0053] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described system, system and unit can refer to the corresponding processes in the foregoing method embodiments, and will not be described here.

[0054] The integrated unit, if implemented in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes a number of instructions for making a propeller fault diagnosis and predictive maintenance device (which can be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the method described in each embodiment of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0055] The above embodiments are only used to illustrate the technical solutions of the present application, rather than limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for fault diagnosis and predictive maintenance of a thruster, characterized in that, The method includes: Step S1: Collect the current signal, temperature signal, vibration signal and pressure signal of the thruster, extract the current harmonic distortion rate, temperature rise rate, vibration peak factor, vibration kurtosis factor and pressure response time, and construct a multi-dimensional time series dataset; Step S2: Map the feature parameters in the multidimensional time series dataset to nodes, establish electromechanical coupling hyperedges, thermal-fluid coupling hyperedges, mechanical degradation hyperedges and sealing failure hyperedges, and construct the hyperedge time series correlation matrix; Step S3: Input the hyperedge temporal correlation matrix into the hypergraph convolutional layer and temporal attention layer of the multiphysics coupled hypergraph temporal encoder for forward propagation calculation, and output the degenerate state feature vector; Step S4: Map the degradation state feature vector to a health index, perform time-series differentiation on the health index to obtain the degradation rate benchmark function, calculate the operating condition acceleration factor based on water depth, speed, load torque and number of start-stop cycles, and obtain the degradation rate function. Step S5: Integrate the reciprocal of the degradation rate function over time. When the health index reaches the failure threshold, terminate the integration to obtain the predicted remaining service life.

2. The method for thruster fault diagnosis and predictive maintenance according to claim 1, characterized in that, Step S1 includes: The three-phase current waveform data of the thruster is collected by a current sensor at a sampling frequency of 1000Hz. The three-phase current waveform data is subjected to Parker transform and Clark transform to extract the d-axis current component and q-axis current component. The total harmonic distortion rate of the d-axis current component and q-axis current component is calculated to obtain the current harmonic distortion rate. Temperature data of the propeller motor windings are collected by a temperature sensor, and the temperature rise rate is obtained by performing a time-series differential operation on the temperature data within a 10-second time window. Radial vibration acceleration data of the magnetic coupling transmission mechanism of the propeller is collected by a vibration accelerometer. The vibration acceleration data is subjected to fast Fourier transform to extract the vibration amplitude at the characteristic frequency. The vibration peak factor is obtained by calculating the ratio of the peak value to the root mean square value of the vibration amplitude. The vibration kurtosis factor is obtained by calculating the ratio of the fourth moment to the square of the second moment of the vibration amplitude. The pressure sensor collects the internal pressure signal and external water pressure signal of the oil-filled sealing chamber of the thruster. The internal pressure signal and the external water pressure signal are differentially calculated to calculate the time interval from the starting point of the pressure difference to the stable end point, and the pressure response time is obtained. The current harmonic distortion rate, the temperature rise rate, the vibration peak factor, the vibration kurtosis factor, and the pressure response time are organized according to a time series, and the water depth, speed, load torque, and cumulative start-stop times at corresponding moments are recorded synchronously to construct the multidimensional time series dataset.

3. The method for thruster fault diagnosis and predictive maintenance according to claim 1, characterized in that, Step S2 includes: The current harmonic distortion rate, temperature rise rate, vibration peak factor, vibration kurtosis factor, and pressure response time in the multidimensional time series dataset are mapped to the first node, the second node, the third node, the fourth node, and the fifth node, respectively, to construct a node set; Establish an electromechanical coupling superedge connecting the first node and the second node, establish a thermal-fluid coupling superedge connecting the second node and the fifth node, establish a mechanical degradation superedge connecting the third node and the fourth node, establish a sealing failure superedge connecting the second node and the fifth node, and construct a superedge set; Construct an association matrix based on the set of nodes and the set of hyperedges, wherein if a node belongs to a hyperedge, the corresponding element of the association matrix is ​​1, otherwise it is 0. A time dimension extension is introduced into the correlation matrix, a preset time window length is set, and each element of the correlation matrix is ​​multiplied by a time weight function, which is an exponential decay function, to obtain the hyperedge temporal correlation matrix.

4. The method for thruster fault diagnosis and predictive maintenance according to claim 1, characterized in that, Step S3 includes: The hyperedge temporal correlation matrix is ​​input into the hypergraph convolutional layer of the multiphysics coupled hypergraph temporal encoder to calculate the node degree matrix and the hyperedge degree matrix. Hypergraph convolution operation is then performed on the hyperedge temporal correlation matrix to obtain the convolutional feature matrix. The convolutional feature matrix is ​​input into the temporal attention layer of the multiphysics coupled hypergraph temporal encoder. The query matrix, key matrix, and value matrix at different times within the time window are calculated. Attention weights are calculated based on the query matrix and the key matrix. The value matrix is ​​weighted and aggregated to obtain the attention feature matrix. The convolutional feature matrix and the attention feature matrix are concatenated and input into the feature fusion layer of the multi-physics coupled hypergraph temporal encoder for dimensionality reduction, and the degenerate state feature vector is output.

5. The method for thruster fault diagnosis and predictive maintenance according to claim 4, characterized in that, The process of inputting the hyperedge temporal correlation matrix into the hypergraph convolutional layer of a multiphysics coupled hypergraph temporal encoder, calculating the node degree matrix and the hyperedge degree matrix, and performing hypergraph convolution operation on the hyperedge temporal correlation matrix to obtain the convolutional feature matrix includes: The hyperedge temporal correlation matrix is ​​input into the hypergraph convolutional layer of the multiphysics coupled hypergraph temporal encoder; Count the number of hyperedges connected to each node in the hyperedge temporal correlation matrix, and construct the diagonal elements of the node degree matrix; Count the number of nodes connected by each hyperedge in the hyperedge temporal correlation matrix, and construct the diagonal elements of the hyperedge degree matrix; The node degree matrix is ​​raised to the power of -1 / 2 to obtain the node degree normalized matrix, and the hyperedge degree matrix is ​​raised to the power of -1 to obtain the hyperedge degree normalized matrix. The node degree normalization matrix, the hyperedge temporal correlation matrix, the hyperedge weight matrix, the hyperedge degree normalization matrix, the transpose of the hyperedge temporal correlation matrix, the node degree normalization matrix, the initial node feature matrix, and the trainable parameter matrix are sequentially multiplied by matrix multiplication. The ReLU activation function is then applied to the result to obtain the convolutional feature matrix.

6. The method for thruster fault diagnosis and predictive maintenance according to claim 1, characterized in that, Step S4 includes: The degraded state feature vector is input into the mapping layer, and the health index is obtained by non-linear mapping through the sigmoid function. Numerical differentiation is performed on the time series of the health indicators to calculate the ratio of the difference between adjacent health indicators to the time interval, thereby obtaining the degradation rate benchmark function. Based on the water depth, speed, load torque, and cumulative number of starts and stops recorded in the multidimensional time series data, the square term of the ratio of water depth to maximum working water depth, the linear term of the ratio of speed to maximum speed, the 1.5 power term of the ratio of load torque to rated torque, and the logarithmic term of the ratio of number of starts and stops to threshold are calculated. The terms are then weighted and summed, and 1 is added to obtain the working condition acceleration factor. The degradation rate reference function is obtained by multiplying the degradation rate reference function by the operating condition acceleration factor.

7. The method for thruster fault diagnosis and predictive maintenance according to claim 1, characterized in that, Step S5 includes: A failure threshold is set to 0.

15. When the health indicator drops to the failure threshold, it is determined that the thruster needs maintenance. Starting from the current moment, the degradation rate function is inversely calculated to obtain the degradation time function; The degradation time function is integrated along the time axis using the trapezoidal rule for numerical integration, with the integration step size set to 1 hour. When the health indicator reaches the failure threshold, the integration calculation is terminated, and the time difference between the time of the integration termination and the current time is used as the predicted value of the remaining service life.

8. A propeller fault diagnosis and predictive maintenance system, characterized in that, For implementing the thruster fault diagnosis and predictive maintenance method as described in any one of claims 1-7, the thruster fault diagnosis and predictive maintenance system comprises: The module is used to collect current, temperature, vibration and pressure signals from the thruster, extract current harmonic distortion rate, temperature rise rate, vibration peak factor, vibration kurtosis factor and pressure response time, and construct a multidimensional time series dataset. The mapping module is used to map the feature parameters in the multidimensional time series dataset to nodes, establish electromechanical coupling hyperedges, thermal-fluid coupling hyperedges, mechanical degradation hyperedges and sealing failure hyperedges, and construct the hyperedge time series correlation matrix; The input module is used to input the hyperedge temporal correlation matrix into the hypergraph convolutional layer and temporal attention layer of the multiphysics coupled hypergraph temporal encoder for forward propagation calculation, and output the degradation state feature vector. The output module is used to map the degradation state feature vector into a health index, perform time-series differentiation on the health index to obtain a degradation rate benchmark function, and calculate the operating condition acceleration factor based on water depth, speed, load torque and number of start-stop cycles to obtain the degradation rate function. The calculation module is used to perform an integral operation on time with respect to the reciprocal of the degradation rate function. When the health indicator reaches the failure threshold, the integration is terminated to obtain the predicted value of the remaining service life.

9. A propeller fault diagnosis and predictive maintenance device, characterized in that, It includes a memory and a processor, the memory storing a computer program that can run on the processor, and the processor executing the computer program to implement the thruster fault diagnosis and predictive maintenance method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is run by the processor, it causes the processor to perform the thruster fault diagnosis and predictive maintenance method as described in any one of claims 1 to 7.

Images