Vibration current characteristics fusion monitoring device for screw pump rotor wear status

Abstract

The present invention discloses a vibration and current feature fusion monitoring device for the wear state of a screw pump rotor, which belongs to the technical field of screw pump monitoring. The device comprises: obtaining the real-time vibration and current signals of the screw pump rotor, constructing a variational mode decomposition objective function with energy constraints, initially dividing the vibration signal into intrinsic mode functions, screening the correlation coefficients, obtaining the vibration subsets of the rotor and hull structure, and then separating the rotor vibration signal by an independent component analysis algorithm. A hierarchical feature fusion strategy is used to synchronize the pre-processed signals, construct a multivariate time series matrix, extract the wear feature set and perform phase space reconstruction to accurately identify sudden wear points. Once identified, the probability of sudden wear is calculated, the wear rate is corrected in combination with the normal wear degree, the remaining life is predicted and the maintenance level is divided. The device aims to solve the problems of existing screw pump monitoring technology being susceptible to interference, delayed response, difficult to adapt to complex working conditions and insufficient model generalization, and to ensure the stable operation of the ship's power system.

Description

Technical Field

[0001] The present invention relates to the technical field of screw pump monitoring, and more particularly to a vibration current feature fusion monitoring device for the wear state of a screw pump rotor. Background Art

[0002] As the intelligentization of ship propulsion systems accelerates, the operational reliability of screw pumps, as core fluid transport equipment, plays a decisive role in navigation safety and operational economics. In actual operation, screw pumps face complex and variable operating conditions, such as variable speeds and loads during navigation, as well as special scenarios like conveying low-temperature or high-viscosity media. Current monitoring technologies for screw pumps have numerous limitations. Most existing technologies rely solely on single vibration or current signals for analysis. For example, methods based on vibration severity thresholds often produce high false alarm rates due to interference from base vibration during ship operation, making it difficult to accurately reflect the true wear status of the screw pump rotor. Monitoring the effective value of motor current not only has a delayed response to initial wear but also fails to effectively capture high-frequency characteristics caused by gap variations, leading to overlooking potential wear risks in the early stages. Furthermore, diagnostic mechanisms based on fixed thresholds cannot flexibly adapt to the dynamic changes in ship operating conditions. Preventive maintenance models based on fixed cycles also exhibit significant drawbacks, easily leading to over- or under-maintenance. This not only wastes resources but also can compromise navigation safety by failing to detect potential faults in a timely manner. The performance of these traditional monitoring technologies is significantly compromised when transporting low-temperature or high-viscosity media. When applying machine learning to screw pump monitoring, traditional models require large amounts of labeled data for training. However, ship screw pump wear data has significant and highly personalized characteristics. This makes traditional models severely lacking in generalization when dealing with new ship types or unexpected operating conditions, making it difficult to accurately monitor and predict the wear status of screw pump rotors.

[0003] In summary, existing screw pump monitoring technology is unable to meet the development needs of intelligent and efficient marine power systems. Therefore, to overcome these limitations, this paper proposes a device that integrates vibration and current characteristics to monitor the wear status of screw pump rotors. This device aims to achieve accurate, real-time monitoring of screw pump rotor wear, providing strong support for the stable operation of marine power systems. Summary of the Invention

[0004] In response to the shortcomings of the existing technology, the purpose of the present invention is to provide a vibration and current feature fusion monitoring device for the wear status of the screw pump rotor. By constructing a series of unique methods such as a variational mode decomposition objective function with energy constraints, a hierarchical feature fusion strategy, and sudden wear probability prediction based on Bayesian networks and Markov chain Monte Carlo algorithms, it is different from the simple vibration monitoring or single feature analysis in the existing technology. It can accurately separate the rotor vibration signal from the vibration and current signals, accurately identify the sudden wear point and starting point, and accurately predict the sudden wear probability and remaining life. It can also scientifically divide the maintenance levels based on this, effectively solving the technical problems of real-time and accurate monitoring of the wear status of the screw pump rotor and maintenance decision-making.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] The vibration current characteristic fusion monitoring device for the wear state of the screw pump rotor includes:

[0007] Real-time data from the screw pump rotor, including vibration and current signals, is acquired. A variational mode decomposition objective function with energy constraints is constructed to initially separate the vibration signal into intrinsic mode functions (IMFs). These IMFs are then screened based on correlation coefficients to obtain rotor and hull structure vibration subsets. The rotor vibration signal is then further separated using an independent component analysis algorithm.

[0008] A hierarchical feature fusion strategy is configured to synchronize the preprocessed rotor vibration and current signals to construct a multivariate time series matrix. Wear features are extracted from the multivariate time series matrix to construct a wear feature set. Sudden wear points are identified by reconstructing the phase space of the wear feature set.

[0009] When a sudden wear point is identified, the sudden wear probability of the screw pump rotor at the sudden wear point is calculated through the Bayesian network and Markov chain. Based on the sudden wear probability and the normal wear degree of the screw pump rotor, the wear rate of the screw pump rotor is corrected to predict the remaining life of the screw pump rotor and divide the maintenance level.

[0010] Specifically, the hierarchical feature fusion strategy includes bottom-level units, middle-level units, and top-level units:

[0011] The bottom unit synchronizes the rotor vibration signal and the current signal with the motor speed pulse as the benchmark, eliminates the time deviation through the interpolation algorithm, and constructs a multivariate time series matrix containing the rotor vibration signal and the current signal;

[0012] The middle-level unit is used to extract wear features from the multivariate time series matrix to construct a wear feature set, and to reconstruct the phase space of the time series of each wear feature in the wear feature set to capture the nonlinear changes of the wear features, obtain the fusion correlation dimension to identify the sudden wear points, and locate the starting point of the sudden wear;

[0013] The top-level unit is used to construct a Bayesian network and a state transition probability matrix through the historical wear feature set of the screw pump rotor when a sudden wear point is identified. The state of the discretized wear feature set from the starting point of the sudden wear to the sudden wear point is used as the initial state of the Markov chain, and the proposal distribution is set. The next candidate state of the Markov chain is generated through the proposal distribution, and the acceptance probability is calculated to obtain the sudden wear probability of the screw pump rotor at the sudden wear point.

[0014] Specifically, the specific steps of separating the rotor vibration signal from the vibration signal include:

[0015] The filter passband is set according to the frequency range of the rotor and hull structure vibration, and a bandpass filter is used to filter and denoise the vibration signal;

[0016] The number of decomposition modes K is set, and based on the variational mode decomposition objective function, the energy entropy constraint term is introduced to construct the variational mode decomposition objective function with energy constraint;

[0017] Lagrange multipliers are introduced to transform the variational mode decomposition objective function with energy constraints into an unconstrained variational problem, and the alternating direction multiplier method is used to solve it to obtain K eigenmode functions, each of which represents a vibration signal with different frequency components.

[0018] Specifically, the specific steps of separating the rotor vibration signal from the vibration signal also include:

[0019] Obtaining rotor vibration reference signals and hull structure vibration reference signals from a signal database, and calculating correlation coefficients between intrinsic mode functions and rotor vibration reference signals and hull structure vibration reference signals respectively;

[0020] According to the size of the correlation coefficient, modal screening is performed to select the rotor vibration subset and the hull structure vibration subset from the K eigenmode functions;

[0021] The screened rotor vibration subset and the hull structure vibration subset are combined into a modal matrix, where each column of the modal matrix represents an intrinsic mode function signal;

[0022] After the modal matrix is ​​mean-centered and whitened, the independent component analysis algorithm is used to further separate the modal matrix into rotor vibration signals and hull structure vibration signals.

[0023] Specifically, the steps of identifying the sudden wear point and locating the starting point of the sudden wear include:

[0024] Set a delay range. Within the delay range, use the mutual information method to calculate the mutual information value of each wear feature in the wear feature set at different delays, select the target delay, and set the embedding dimension.

[0025] For the time series of each wear feature in the wear feature set, phase space reconstruction is performed;

[0026] For each wear feature phase space, the neighborhood radius is set according to the statistical value of the phase space vector distance. For each pair of phase space vectors in the phase space, the distance between the phase space vectors is calculated to calculate the correlation integral of the wear feature.

[0027] The correlation integral is logarithmically transformed and linearly fitted to obtain the slope of the fitting line as the correlation dimension. The wear feature importance scores of the wear feature set are weighted and normalized to assign weights to each wear feature to obtain the fusion correlation dimension.

[0028] Specifically, the steps of identifying the sudden wear point and locating the sudden wear starting point also include:

[0029] Based on the historical data of the screw pump rotor, a correlation dimension threshold is configured. If the fused correlation dimension is greater than the correlation dimension threshold, it is determined that a sudden wear point exists and is marked.

[0030] Set a backtracking cycle to form a backtracking window. Within the backtracking window, perform a backtracking search starting from the sudden wear point. Calculate the relative change rate of the fusion correlation dimension at each time point to calculate the path cost at each time point within the backtracking window and obtain the path cost curve within the backtracking window.

[0031] Calculate the slope of the path cost curve at each time point in the backtracking window, filter the slope mutation points in the backtracking window, and sort them in chronological order. The time point where the slope mutation point is far away from the sudden wear point is taken as the starting point of the sudden wear.

[0032] Specifically, the specific steps for predicting the probability of sudden wear of the screw pump rotor include:

[0033] Obtain the historical wear feature set of the screw pump rotor and mark the actual wear state corresponding to each data sample; discretize the historical wear feature set and divide the state interval for each wear feature;

[0034] A Bayesian network is constructed to determine the causal relationship between each node in the Bayesian network. The conditional probability distribution of each node under different state combinations of its parent node is obtained using the historical wear feature set of the screw pump rotor.

[0035] Based on the discretization results of the historical wear feature set, the number of state transitions is determined. Based on the Bayesian network, the probability of transitioning from one state to another is calculated, thereby constructing a state transition probability matrix.

[0036] The wear feature set from the sudden wear starting point to the sudden wear point is obtained and discretized, and the state divided according to the discretization of the wear feature set from the sudden wear starting point to the sudden wear point is used as the initial state of the Markov chain.

[0037] Specifically, the specific steps of predicting the probability of sudden wear of the screw pump rotor also include:

[0038] According to the state of the discretized wear feature set at each time point from the start point of sudden wear to the sudden wear point, a proposal distribution is set to generate the next candidate state of the Markov chain;

[0039] After generating the next candidate state, the joint probability of the current state and the candidate state in the state space is calculated based on the Bayesian network and the conditional probability distribution of each node. Based on the proposal distribution, the probability of transitioning from the candidate state to the current state and the probability of transitioning from the current state to the candidate state are calculated respectively to calculate the acceptance probability.

[0040] Generate a random number that obeys a uniform distribution. If the random number is less than or equal to the acceptance probability, accept the candidate state and use it as the next state of the Markov chain; otherwise, reject the candidate state and keep the current state.

[0041] Iterative sampling is performed until the Markov chain converges. The number of samples in the sudden wear state and the total number of sampling times are counted. The ratio of the number of samples in the sudden wear state to the total number of sampling times is calculated as the sudden wear probability of the screw pump rotor at the sudden wear point.

[0042] Specifically, the steps to construct a multivariate time series matrix include:

[0043] Obtain the speed pulse signal of the motor shaft as the physical synchronization reference for the rotor vibration signal and current signal, generate a timestamp mark, and establish a synchronous clock reference coordinate system;

[0044] Performing time stamps on the rotor vibration signal and the current signal, and mapping the time stamps of the rotor vibration signal and the current signal to a synchronous clock reference coordinate system;

[0045] Determine whether to trigger interpolation correction based on the accumulated timestamp deviation between the rotor vibration signal and the current signal;

[0046] If the interpolation correction is triggered, the following steps are performed: the curvature value of each sampling point of the rotor vibration signal and the current signal is calculated, the local curvature extreme value points are selected as interpolation candidate nodes, and an interpolation node sequence is selected from the interpolation candidate nodes;

[0047] Based on the interpolation node sequence, the corresponding timestamps of the rotor vibration signal and the current signal are obtained respectively, and the time offset between the interpolation nodes at the same position of the rotor vibration signal and the current signal is calculated;

[0048] The calculated time offset is used to construct a time deviation matrix to obtain the nonlinear mapping relationship between the rotor vibration signal and the current signal;

[0049] Based on the current signal time series, the interpolation time points of the rotor vibration signal are obtained through nonlinear mapping, and the aligned rotor vibration signal is generated using the interpolation algorithm.

[0050] Based on the rotor vibration signal time series, the interpolation time points of the current signal are obtained through nonlinear mapping, and the interpolation algorithm is used to generate the aligned current signal.

[0051] The aligned rotor vibration signals and current signals are arranged in time sequence to generate a spatiotemporally aligned multivariate time series matrix.

[0052] Specifically, the steps for classifying maintenance levels include:

[0053] When a sudden wear point is identified, the sudden wear probability of the screw pump rotor at the sudden wear point is obtained, and the wear degree per unit time is obtained based on the historical wear data of the screw pump rotor during normal operation;

[0054] Set a correction factor to correct the wear rate per unit time according to the probability of sudden wear to obtain the corrected wear rate;

[0055] Determine the wear limit of the screw pump rotor when it reaches scrapping or requires major repairs, obtain the cumulative usage time of the screw pump rotor, the number of sudden wear points, and the probability of sudden wear of the screw pump rotor at the sudden wear points, so as to obtain the current wear amount of the screw pump rotor and use the corrected wear rate to predict the remaining life;

[0056] Maintenance thresholds are set, including an upper maintenance threshold and a lower maintenance threshold, and the screw pump rotor is divided into different maintenance levels according to the predicted remaining life.

[0057] Beneficial effects of the present invention:

[0058] By fusing the vibration and current signals and constructing a variational mode decomposition objective function with energy constraints, combined with correlation coefficient screening and independent component analysis algorithms, the rotor vibration signal can be accurately separated from the complex vibration signal. This process effectively shields the vibration interference of the hull structure, greatly improving the accuracy of signal analysis and significantly reducing the false alarm rate. In the hierarchical feature fusion strategy, the bottom unit uses the motor speed pulse as the benchmark and uses the interpolation algorithm to construct a multivariate time series matrix to ensure signal synchronization; the middle unit extracts features and identifies sudden wear points through phase space reconstruction; the top unit calculates the probability of sudden wear by constructing a Bayesian network and a Markov chain. The close collaboration between the various units enables a comprehensive and in-depth analysis of the wear status, timely and accurate capture of sudden wear points and starting points, and makes up for the shortcomings of the delayed response of traditional technologies.

[0059] Throughout the entire monitoring process, from signal processing and feature extraction to probability prediction, a complete and efficient technology chain is formed. Ultimately, this enables accurate prediction of the probability of sudden wear and the remaining lifespan, rationally categorizing maintenance levels, avoiding resource waste due to excessive maintenance and safety hazards caused by insufficient maintenance, and effectively ensuring the stable and efficient operation of the ship's power system under complex operating conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 Schematic diagram of the logic structure of the vibration current feature fusion monitoring device for the wear state of the screw pump rotor of the present invention;

[0061] Figure 2 This is a flow chart of the specific steps of separating the rotor vibration signal and the hull structure vibration signal according to the present invention;

[0062] Figure 3 Schematic diagram of the hierarchical feature fusion strategy of the present invention;

[0063] Figure 4 A flowchart of the specific steps for constructing a multivariate time series matrix of the present invention;

[0064] Figure 5 A flowchart of the specific steps of identifying sudden wear points and locating the starting point of sudden wear according to the present invention;

[0065] Figure 6 A flowchart of the specific steps for generating maintenance decisions for the present invention. DETAILED DESCRIPTION

[0066] See also Figure 1 ,This embodiment introduces a vibration current feature fusion monitoring device for the wear state of a screw pump rotor, including a signal acquisition module, a feature processing module, a maintenance decision module and a user interaction module;

[0067] The signal acquisition module is used to collect real-time data through a sensor array, including vibration signals and current signals. Specifically, a three-axis accelerometer is arranged along the axial direction of the screw pump stator housing to collect vibration signals. The rotor vibration signal and the hull structure vibration signal are separated using variational mode decomposition and independent component analysis algorithms, and the screw pump motor current signal is simultaneously collected.

[0068] In this embodiment, three-axis accelerometers are evenly arranged along the axial direction of the stator housing of the screw pump. In order to ensure that the sensor is firmly installed and does not affect the normal operation of the pump, a magnetic installation method is adopted to fix the sensor on the surface of the stator housing. At the same time, the installation position of each sensor is numbered and marked to facilitate subsequent data processing and analysis. The three-axis accelerometer collects the vibration signal of the stator housing of the screw pump in real time and captures the high-frequency vibration signal according to the set collection frequency. The collected vibration signal is a mixed signal of the vibration signal of the screw pump rotor and the vibration signal of the hull structure. Wrap the Rogowski coil around the input phase line of the screw pump motor and fix it with a special fixing clamp to ensure that the coil fits tightly with the phase line to collect the current signal.

[0069] See also Figure 2 Preferably, the specific steps of separating the rotor vibration signal and the hull structure vibration signal include:

[0070] A bandpass filter is used to filter and de-noise the vibration signal, with the passband set based on the frequency range of the rotor and hull structure vibrations. This filter removes high-frequency noise and low-frequency interference from the vibration signal. By analyzing the vibration characteristics of the rotor and hull structure under normal operating conditions, the vibration frequency range of each is determined. Based on this range, the passband range of the bandpass filter is set, allowing vibration signals within this range to pass smoothly while effectively filtering out high-frequency noise and low-frequency interference.

[0071] The decomposition mode number K is set according to the experimental data. On the basis of the traditional variational mode decomposition objective function, the energy entropy constraint term is introduced to construct the variational mode decomposition objective function with energy constraint, namely:

[0072] ;

[0073] in, represents minimizing the sum of the estimated bandwidths of each eigenmode function, where is the unit impulse function, is an imaginary unit, is the time variable, is the time variable The partial derivative of is the convolution operator, is the complex demodulation factor, is the center frequency of the kth mode, which is used to characterize the characteristic frequency of the mode. is the kth eigenmode function, representing the vibration signal of a certain frequency component, is a penalty term used to balance signal fidelity and modal bandwidth. is the penalty factor, ranging from [2000, 5000], adjusted according to the complexity of the signal. is the original vibration signal, is the energy entropy of the kth mode, which is used to measure the uniformity of the energy distribution of the mode. It is the constraint weight, which is used to adjust the influence of the energy entropy constraint term, and its value range is [0.1, 0.5].

[0074] Lagrange multipliers are introduced to transform the variational modal decomposition objective function with energy constraints into an unconstrained variational problem, and the alternating direction multiplier method is used to solve it to obtain K intrinsic mode functions, each of which represents a vibration signal with different frequency components. This decomposition method that introduces energy entropy constraints can better consider the energy distribution characteristics of the signal and improve the accuracy and rationality of the decomposition.

[0075] Obtain rotor vibration reference signals and hull structure vibration reference signals from the signal database. The reference signals are obtained by respectively collecting rotor vibration alone and hull vibration alone under an ideal experimental environment. Calculate the correlation coefficients between the intrinsic mode function and the rotor vibration reference signals and the hull structure vibration reference signals respectively.

[0076] Based on the magnitude of the correlation coefficient, modal screening is performed to select rotor vibration subsets and hull structure vibration subsets from the K eigenmode functions. Specifically, based on the magnitude of the correlation coefficient, the eigenmode functions with the highest correlation are classified as corresponding vibration types. For example, if the correlation coefficient of a particular eigenmode function with the rotor vibration reference signal is greater than the correlation coefficient with the hull structure vibration reference signal, and the correlation coefficient is greater than a set correlation threshold, then the eigenmode function is classified as an eigenmode function related to rotor vibration; otherwise, it is classified as an eigenmode function related to hull structure vibration. Ultimately, the rotor vibration subset and the hull structure vibration subset related to the rotor vibration and hull structure vibration are selected.

[0077] The filtered rotor vibration subsets and the hull structure vibration subsets are combined into a modal matrix. Each column of the modal matrix represents an intrinsic mode function signal. The filtered signals are then integrated to form an input data structure suitable for independent component analysis (ICA). This matrix representation effectively organizes signal information, facilitating further signal separation by the ICA algorithm.

[0078] After mean centering and whitening the modal matrix, the independent component analysis algorithm is used to further separate the modal matrix. Based on the signal independence assumption, the separation matrix is ​​continuously adjusted through iterative calculations. After multiple iterations, the modal matrix is ​​finally separated into the independent rotor vibration signals and hull structure vibration signals. This separation method exploits the statistical independence between the signals and can effectively separate the mixed different source signals.

[0079] The feature processing module performs data preprocessing on the rotor vibration and current signals. Specifically, it applies adaptive noise cancellation technology to the rotor vibration signals separated by the signal acquisition module to effectively remove interference noise. Discrete wavelet transform is used to reduce noise on the collected current signals, and Park transform is then used to extract the DC component of the current signal for clearer analysis of the current signal characteristics. A hierarchical feature fusion strategy is configured to perform baseline alignment on the preprocessed rotor vibration and current signals, extract wear features, and then identify potential wear points and predict potential wear probability.

[0080] See also Figure 3 ,Preferably, the hierarchical feature fusion strategy includes bottom-level units, middle-level units and top-level units:

[0081] The bottom-level unit is used to construct a multivariate time series matrix that is aligned in time and space. The rotor vibration signal and current signal are synchronized based on the motor speed pulse. The interpolation algorithm is used to eliminate time deviation to construct a multivariate time series matrix containing the rotor vibration signal and current signal, ensuring the temporal and spatial consistency of the data.

[0082] See also Figure 4 Specifically, the steps for constructing a multivariate time series matrix include:

[0083] The photoelectric sensor configured on the screw pump rotor motor shaft obtains the motor shaft speed pulse signal, generating one synchronization pulse per revolution as the physical synchronization reference for the rotor vibration signal and current signal, and generates a timestamp mark to establish a synchronous clock reference coordinate system.

[0084] Obtain the timestamp marks of the rotor vibration signal and current signal sampling points respectively, and calculate the timestamp deviation of the rotor vibration signal and current signal in real time;

[0085] Configure the time deviation threshold to accumulate the timestamp deviation between the rotor vibration signal and the current signal in real time. If the timestamp deviation is greater than the time deviation threshold, interpolation correction is triggered.

[0086] If the interpolation correction is not triggered, the rotor vibration signal and the current signal are arranged in time sequence to generate a time-space aligned multivariate time series matrix:

[0087] If interpolation correction is triggered, then:

[0088] Calculate the curvature value of each sampling point of the rotor vibration signal and current signal, and mark the local curvature extreme points. These points usually correspond to the locations where the signal suddenly changes or changes dramatically, and serve as candidate interpolation nodes;

[0089] A set of interpolation node sequences is selected from the interpolation candidate nodes by a dynamic programming algorithm, so that the curvature change rate between adjacent nodes in the interpolation node sequence is less than a preset curvature threshold, and the number of nodes in the interpolation node sequence is the least;

[0090] For the interpolation node sequences of the rotor vibration signal and the current signal, the corresponding timestamps are obtained respectively, and the time offset between the interpolation nodes of the rotor vibration signal and the current signal at the same position is calculated;

[0091] The calculated time offset is used to construct a time deviation matrix, where the number of rows is the number of nodes of the rotor vibration signal and the number of columns is the number of nodes of the current signal.

[0092] Using methods such as polynomial fitting or neural networks, a nonlinear mapping relationship between vibration signal time and current signal time is constructed based on the time deviation matrix;

[0093] Based on the current signal time series, the interpolation time points of the rotor vibration signal are obtained through nonlinear mapping, and the aligned rotor vibration signal is generated using the interpolation algorithm.

[0094] Based on the rotor vibration signal time series, the interpolation time points of the current signal are obtained through nonlinear mapping, and the interpolation algorithm is used to generate the aligned current signal.

[0095] The aligned rotor vibration signals and current signals are arranged in time sequence to generate a spatiotemporally aligned multivariate time series matrix.

[0096] The middle-level unit is used to extract wear features from the multivariate time series matrix to construct a wear feature set, and to reconstruct the phase space of the time series of each wear feature in the wear feature set to capture the nonlinear changes of the wear features, obtain the fusion correlation dimension to identify the sudden wear points, and locate the starting point of the sudden wear;

[0097] In this embodiment, after data cleaning, data outlier processing, and data standardization are performed on the multivariate time series matrix generated by the bottom unit, wear feature extraction is performed. Wear features with high correlation with the wear state are selected through methods such as correlation analysis and variance analysis. The wear features are then reduced in dimension using principal component analysis or linear discriminant analysis to obtain the final wear feature set.

[0098] Specifically, wear feature extraction includes:

[0099] Extract time domain features of multivariate time series, including signal kurtosis, which reflects the sharpness of the signal; signal kurtosis, which measures the prominence of the signal peak; root mean square value, which reflects the energy of the signal; waveform index, which reflects the change of the signal relative to the average;

[0100] Extract the frequency domain characteristics of the multivariate time series, perform frequency domain changes on the multivariate time series to obtain the frequency domain signal, extract the amplitude of different frequency components in the frequency domain signal to reflect the energy distribution of the frequency domain signal at each frequency, and extract the phase information of the frequency components to analyze the relative position and synchronization of the frequency domain signal, calculate the frequency center of gravity of the frequency domain signal, and reflect the main frequency distribution of the frequency domain signal.

[0101] See also Figure 5 Preferably, the specific steps of identifying the sudden wear point and locating the sudden wear starting point include:

[0102] According to the frequency range of the screw pump, the delay range is set. Within the delay range, the mutual information method is used to calculate the mutual information value of each wear feature in the wear feature set at different delays. Mutual information is used to measure the degree of dependence between two random variables. In the monitoring of the screw pump, for each wear feature in the wear feature set and its delayed wear feature, the mutual information reflects the correlation of the wear features at different delays. When the mutual information value is large, it means that the correlation of the wear features at this delay is strong; when the mutual information value is small, it means that the dependence between the wear features is weak. At this time, the signal may contain more new information. According to the size of the mutual information value, the target delay is selected. The target delay is typically selected based on the minimum mutual information value. When the mutual information value reaches a minimum, it means that the wear characteristics at that delay are least correlated with the original wear characteristics, containing more independent information and more suitable for phase space reconstruction.

[0103] Set the embedding dimension , embedding dimension It can be determined by the Cao method. By calculating the Cao index under different embedding dimensions, when the index tends to be stable, the minimum embedding dimension is obtained as the embedding dimension. This avoids information loss or redundancy caused by improper dimension selection and provides a suitable spatial structure for subsequent accurate analysis of changes in wear characteristics.

[0104] For the time series of each wear feature in the wear feature set, phase space reconstruction is performed. , the phase space reconstruction includes:

[0105] ;

[0106] in, After reconstruction, indivual dimensional phase space vector, Is a wear characteristic The time series of The original data points, is the target delay, is the embedding dimension, The value range is , is the length of the time series; it enables the changes in wear characteristics to be presented in the phase space in a more intuitive and easier to analyze manner, providing an effective data representation for further analysis of the dynamic behavior of wear characteristics.

[0107] For each wear feature phase space, the neighborhood radius is set according to the statistical value of the phase space vector distance, including the mean and median. For example, the neighborhood radius is set to one tenth of the median. For each pair of phase space vectors in the phase space, the distance between the phase space vectors is calculated to calculate the correlation integral of the wear characteristics, that is:

[0108] ;

[0109] in, The neighborhood radius is The correlation integral of the wear characteristics when It is and The distance between the phase space vectors, is a step function, when hour, ,when hour, ; Quantify the degree of correlation between vectors in the phase space, thereby reflecting the distribution and mutual relationship of wear characteristics in the phase space.

[0110] The correlation integral is logarithmically transformed and linearly fitted. The slope of the fitted line is obtained as the correlation dimension. The wear feature importance scores of the wear feature set are weighted and normalized to assign weights to each wear feature to obtain the fused correlation dimension. The wear feature importance scores can be obtained from the eigenvalues ​​of the feature covariance matrix of the principal component analysis or the ratio of the between-class scatter matrix to the within-class scatter matrix of the linear discriminant method.

[0111] Based on the historical data of the screw pump rotor, a correlation dimension threshold is configured. If the fused correlation dimension exceeds the threshold, a sudden wear point is determined and marked. This allows for timely detection of abnormal wear during screw pump operation, providing important early warning information for subsequent action, helping to avoid equipment failures and downtime losses caused by wear. If the fused correlation dimension is not greater than the threshold, no action is taken.

[0112] Set the backtracking cycle to form a backtracking window. Within the backtracking window, perform a backtracking search starting from the sudden wear point. Calculate the relative change rate of the fusion correlation dimension at each time point to calculate the path cost at each time point within the backtracking window and obtain the path cost curve within the backtracking window. The calculation formula for the path cost is as follows:

[0113] ;

[0114] in, is the time point of sudden wear. is the time point within the lookback window The path cost, Is a positive integer used to determine the time point in the lookback window relative to the sudden wear point time point position offset. Indicates the time point within the lookback window from the sudden wear point have By changing the time point of the time interval The value of can be used to calculate the path cost at different time points in the backtracking window. It's time By calculating the relative rate of change of the fused correlation dimension at each time point and computing the path cost, we can quantify the cumulative degree of wear characteristic change at each time point prior to the sudden wear point. Obtaining the path cost curve within the lookback window intuitively demonstrates the cumulative trend of wear characteristic change over time, providing an important basis for further analysis of the onset of sudden wear.

[0115] Calculating the slope of the path cost curve at each time point within the lookback window reflects the rate of change of the path cost over time. Slope mutation points within the lookback window are screened and sorted chronologically, with the time point at which the slope mutation point, farthest from the sudden wear point, being designated as the sudden wear starting point. Identifying the time point when wear characteristics begin to intensify from the changes in the path cost curve provides a precise time location for a deeper understanding of the onset and development of sudden wear, helping to more accurately analyze the causes and mechanisms of wear and provide more targeted recommendations for equipment maintenance and management.

[0116] The top-level unit is used to construct a state transition probability matrix through the Bayesian network when a sudden wear point is identified, and combine it with the Markov chain Monte Carlo algorithm to predict the sudden wear probability of the screw pump rotor at the sudden wear point;

[0117] Preferably, the specific steps of predicting the probability of sudden wear of the screw pump rotor include:

[0118] Obtain the historical wear feature set of the screw pump rotor, including the wear feature set of the screw pump under normal and abnormal operating conditions, and mark the actual wear state corresponding to each data sample, that is, whether sudden wear occurs;

[0119] Discretize the historical wear feature set, and for each wear feature, divide the state interval. For example, the kurtosis value of the vibration signal is divided into low kurtosis, medium kurtosis, and high kurtosis; the speed is divided into low speed, medium speed, and high speed. This step prepares for the subsequent construction of the conditional probability distribution of the Bayesian network.

[0120] Based on the operating principle of the screw pump, a Bayesian network was constructed to identify the causal relationships between the nodes within the Bayesian network and construct a directed acyclic graph. For example, the speed of the screw pump affects the vibration and current signal characteristics, which in turn affect the wear state. Therefore, directed edges exist from the speed node to the vibration and current signal characteristic nodes, and from the vibration and current signal characteristic nodes to the wear state node.

[0121] Using the historical wear feature set of screw pump rotors, we use maximum likelihood estimation or Bayesian estimation to obtain the conditional probability distribution of each node under different combinations of its parent node states. For example, for a node in the wear state, we determine the probability of it being in a state of sudden wear, normal wear, etc., given different combinations of vibration and current signal features.

[0122] Based on the discretization results of the historical wear feature set, the number of state transitions is determined. Based on the Bayesian network, the probability of transitioning from one state to another is calculated. That is, for each pair of states (p, q), the probability of transitioning to state q at the next moment when the current state is p is calculated, thereby constructing a state transition probability matrix.

[0123] Obtain the wear feature set from the start point of sudden wear to the sudden wear point, and discretize it. The state divided by the discretization of the wear feature set from the start point of sudden wear to the sudden wear point is used as the initial state of the Markov chain.

[0124] According to the discretized state of the wear feature set at each time point from the start point of sudden wear to the sudden wear point, a proposal distribution is set to generate the next candidate state of the Markov chain. The proposal distribution can be designed as a random perturbation pattern based on the current state. For example, for the state of each wear feature, transitions are made between its adjacent states with a certain probability, or small random adjustments are made to the state combination of multiple wear features.

[0125] After generating the next candidate state, calculate the acceptance probability ,Right now:

[0126] ;

[0127] in, and are the probabilities of the current state and the next candidate state under the state transition probability matrix determined by the Bayesian network, which are accurately calculated through the joint probability distribution of the Bayesian network; and is the probability of the proposed distribution, indicating the probability of Transfer to current state , and from the current state Transfer to Candidate Status probability.

[0128] Generate a random number that follows a uniform distribution. If the random number is less than or equal to the acceptance probability, then accept the candidate state. , take it as the next state of the Markov chain; otherwise keep the current state;

[0129] Repeat the above steps of generating candidate states, calculating acceptance probabilities, and accepting or rejecting candidate states, and perform iterative sampling until the Markov chain converges to a stationary distribution;

[0130] After the Markov chain converges, the number of samples in the sudden wear state and the total sampling times are counted, the ratio of the number of samples in the sudden wear state to the total sampling times is calculated, and the sudden wear probability of the screw pump rotor at the sudden wear point is calculated.

[0131] The maintenance decision module is used to correct the wear rate of the screw pump rotor based on the predicted probability of sudden wear and the normal wear of the screw pump rotor, and predict the remaining life of the screw pump rotor to divide the maintenance level;

[0132] See also Figure 6 , preferably, the specific steps of generating a maintenance decision include:

[0133] When a sudden wear point is identified, the sudden wear probability of the screw pump rotor at the sudden wear point is obtained;

[0134] Obtain historical wear data of the screw pump rotor during normal operation and obtain the wear degree per unit time under normal use , can be obtained by measuring the wear amount at different time points and then calculating the ratio of the difference to the time interval;

[0135] Setting the correction factor The wear degree per unit time is corrected according to the probability of sudden wear. The correction factor can be determined by establishing a function related to the probability of sudden wear to adjust the degree of influence of the probability of sudden wear on the wear rate. The correction factor value can also be determined by analyzing the correlation between sudden wear events and wear rate changes in historical data.

[0136] According to the correction factor, the corrected wear rate is obtained ,When the probability of sudden wear is high, the wear rate will increase accordingly to more accurately reflect the actual wear conditions faced by the equipment.

[0137] Based on the design specifications and safe operation standards of the screw pump rotor, determine the wear limit value of the rotor when it reaches scrapping or requires major repairs;

[0138] Obtain the cumulative usage time of the screw pump rotor, the number of sudden wear points, and the sudden wear probability of the screw pump rotor at the sudden wear points to obtain the current wear amount of the screw pump rotor. Use the corrected wear rate to predict the remaining life, that is:

[0139] ;

[0140] in, is the predicted remaining lifespan, is the wear limit value, is the current wear of the screw pump rotor, is the corrected wear rate;

[0141] Set maintenance thresholds, including upper and lower thresholds, and classify the screw pump rotor into different maintenance levels based on the predicted remaining life. For example, when the predicted remaining life is greater than the upper threshold, it is considered a low maintenance level, and routine inspections and maintenance measures can be implemented, such as regularly checking equipment operating parameters and cleaning the equipment. When the predicted remaining life is less than or equal to the upper threshold and greater than the lower threshold, it is considered a medium maintenance level. In addition to routine inspections, the frequency of key component inspections needs to be increased, parts that may need replacement should be prepared, and consideration should be given to arranging training for maintenance personnel in advance. When the maintenance level is less than or equal to the lower threshold, it is considered a high maintenance level, and a comprehensive shutdown inspection should be immediately arranged. A detailed maintenance plan should be developed, including replacing severely worn parts and conducting a comprehensive overhaul and commissioning of the equipment.

[0142] During screw pump operation, actual wear is regularly measured, wear conditions are tracked in real time, and the probability of sudden wear is regularly recalculated, wear rates are corrected, remaining life is predicted, and maintenance decisions are updated accordingly. If any anomalies are detected during monitoring, maintenance decisions should be immediately reassessed and adjusted to ensure the equipment remains in a safe and reliable operating state.

[0143] The user interaction module displays the remaining life trend curve and maintenance level of the screw pump rotor in an intuitive graphical format. Color coding is used to intuitively reflect the status of the screw pump rotor. When the screw pump rotor status is in the red zone, it indicates that the equipment is at serious risk of wear and requires immediate maintenance; when it is in the yellow zone, it indicates that the equipment has a certain risk of wear and requires close attention; when it is in the green zone, it indicates that the equipment is operating normally.

[0144] Working principle and its effect:

[0145] Sensors are used to collect real-time vibration and current signals from the screw pump rotor. The vibration signals are first filtered and denoised using a bandpass filter based on the vibration frequency ranges of the rotor and hull structure. Next, the number of decomposed modes is set, and an energy entropy constraint term is introduced into the traditional variational modal decomposition objective function. A new objective function is constructed, which is converted into an unconstrained variational problem using the Lagrange multiplier method and solved using the alternating direction multiplier method to obtain K eigenmode functions. The correlation coefficients of these functions are then calculated with the rotor and hull structure vibration reference signals obtained from the signal database. The rotor vibration subset and the hull structure vibration subset are screened and combined into a modal matrix. After mean centering and whitening, the pure rotor vibration signal is further separated using the independent component analysis algorithm.

[0146] A hierarchical feature fusion strategy is configured. The bottom-level unit uses the motor speed pulse as a reference to timestamp the rotor vibration and current signals, mapping them to a synchronous clock reference coordinate system. The accumulated timestamp deviation determines whether to trigger interpolation correction. Interpolation candidate nodes are screened by calculating curvature values. A time deviation matrix is ​​constructed to obtain nonlinear mapping relationships, generating a multivariate time series matrix aligned in time and space. The middle-level unit extracts wear features from this matrix, constructs a wear feature set, and reconstructs the phase space of each wear feature time series. The correlation integral is calculated by setting a neighborhood radius. The correlation dimension is obtained through logarithmic transformation and linear fitting. The fused correlation dimension is then normalized using the wear feature importance scores to identify sudden wear points and locate their starting points.

[0147] After identifying a sudden wear point, the top-level unit obtains the historical wear signature set of the screw pump rotor, constructs a Bayesian network, determines the causal relationships between nodes, obtains a conditional probability distribution, and constructs a state transition probability matrix. Using the discretized state of the sudden wear starting point as the initial state of the Markov chain, a proposal distribution is set to generate candidate states. Joint and acceptance probabilities are calculated using the Bayesian network. After iterative sampling until the Markov chain converges, the probability of sudden wear is calculated. The wear rate is corrected based on normal wear. The current wear amount is calculated based on the wear limit and accumulated usage time, and the remaining life is predicted. Maintenance thresholds are set to categorize maintenance levels.

[0148] This invention effectively eliminates interference from hull structure vibrations and accurately captures rotor vibration signals, laying a solid foundation for reliable data for subsequent in-depth analysis and significantly reducing the false alarm rate of monitoring. The hierarchical feature fusion strategy employed can comprehensively capture wear characteristics, accurately identify sudden wear points and their starting points, and thoroughly resolve the problem of delayed response from traditional technologies, providing early warnings of potential faults. By accurately predicting the probability of sudden wear and remaining lifespan, and rationally categorizing maintenance levels, over- or under-maintenance can be avoided, effectively ensuring the stable operation of screw pumps, effectively reducing ship operating costs, and significantly improving navigation safety.

[0149] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiment. All technical solutions based on the concept of the present invention are within the scope of protection of the present invention. It should be noted that for those skilled in the art, various improvements and modifications that do not depart from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A vibration current characteristic fusion monitoring device for the wear state of a screw pump rotor, characterized in that: include: Real-time data of the screw pump rotor, including vibration and current signals, is acquired. A variational mode decomposition objective function with energy constraints is constructed to preliminarily separate the vibration signals into intrinsic mode functions (IMFs). The IMFs are screened based on correlation coefficients to obtain rotor vibration subsets and hull structure vibration subsets. The rotor vibration signal is then separated using an independent component analysis algorithm. configuring a hierarchical feature fusion strategy, performing data synchronization processing on the preprocessed rotor vibration signal and the current signal through the hierarchical feature fusion strategy to construct a multivariate time series matrix; extracting wear features from the multivariate time series matrix to construct a wear feature set, and identifying sudden wear points by performing phase space reconstruction on the wear feature set; When the sudden wear point is identified, the sudden wear probability of the screw pump rotor at the sudden wear point is calculated using a Bayesian network and a Markov chain; based on the sudden wear probability and the normal wear degree of the screw pump rotor, the wear rate of the screw pump rotor is corrected to predict the remaining life of the screw pump rotor and classify the maintenance level; The hierarchical feature fusion strategy includes bottom-level units, middle-level units, and top-level units: The bottom layer unit synchronizes the rotor vibration signal and the current signal with the motor speed pulse as the reference, eliminates the time deviation through the interpolation algorithm, and constructs a multivariate time series matrix containing the rotor vibration signal and the current signal; The middle-level unit is used to extract wear features from the multivariate time series matrix to construct a wear feature set, and to perform phase space reconstruction on the time series of each wear feature in the wear feature set to capture the nonlinear changes of the wear features, obtain the fused correlation dimension to identify sudden wear points, and locate the starting point of sudden wear; The top-level unit is used to construct a Bayesian network and a state transition probability matrix based on the historical wear feature set of the screw pump rotor when a sudden wear point is identified, and use the state of the discretized wear feature set from the sudden wear starting point to the sudden wear point as the initial state of the Markov chain, set a proposal distribution, generate the next candidate state of the Markov chain based on the proposal distribution, and calculate the acceptance probability to obtain the sudden wear probability of the screw pump rotor at the sudden wear point; The specific steps of predicting the probability of sudden wear of the screw pump rotor include: Obtain the historical wear feature set of the screw pump rotor and mark the actual wear state corresponding to each data sample; discretize the historical wear feature set and divide the state interval for each wear feature; A Bayesian network is constructed to determine the causal relationship between each node in the Bayesian network. The conditional probability distribution of each node under different state combinations of its parent node is obtained using the historical wear feature set of the screw pump rotor. Based on the discretization results of the historical wear feature set, the number of state transitions is determined. Based on the Bayesian network, the probability of transitioning from one state to another is calculated, thereby constructing a state transition probability matrix. Obtain the wear feature set from the start point of sudden wear to the sudden wear point, and discretize it. The state divided by the discretization of the wear feature set from the start point of sudden wear to the sudden wear point is used as the initial state of the Markov chain. According to the state of the discretized wear feature set at each time point from the start point of sudden wear to the sudden wear point, a proposal distribution is set to generate the next candidate state of the Markov chain; After generating the next candidate state, the joint probability of the current state and the candidate state in the state space is calculated based on the Bayesian network and the conditional probability distribution of each node. Based on the proposal distribution, the probability of transitioning from the candidate state to the current state and the probability of transitioning from the current state to the candidate state are calculated respectively to calculate the acceptance probability. Generate a random number that obeys a uniform distribution. If the random number is less than or equal to the acceptance probability, accept the candidate state and use it as the next state of the Markov chain; otherwise, reject the candidate state and keep the current state. Iterative sampling is performed until the Markov chain converges. The number of samples in the sudden wear state and the total number of sampling times are counted. The ratio of the number of samples in the sudden wear state to the total number of sampling times is calculated as the sudden wear probability of the screw pump rotor at the sudden wear point.

2. The vibration current characteristic fusion monitoring device for the wear state of the screw pump rotor according to claim 1, characterized in that: The specific steps of separating the rotor vibration signal from the vibration signal include: Set the filter passband and use a bandpass filter to filter and denoise the vibration signal; The number of decomposition modes K is set, and based on the variational mode decomposition objective function, the energy entropy constraint term is introduced to construct the variational mode decomposition objective function with energy constraint; Lagrange multipliers are introduced to transform the variational mode decomposition objective function with energy constraints into an unconstrained variational problem, and the alternating direction multiplier method is used to solve it to obtain K intrinsic mode functions, each of which represents a vibration signal with different frequency components.

3. The vibration current characteristic fusion monitoring device for the wear state of the screw pump rotor according to claim 2, characterized in that: The specific step of separating the rotor vibration signal from the vibration signal also includes: Obtaining rotor vibration reference signals and hull structure vibration reference signals from a signal database, and calculating correlation coefficients between intrinsic mode functions and rotor vibration reference signals and hull structure vibration reference signals respectively; Perform modal screening based on the correlation coefficient and select the rotor vibration subset and the hull structure vibration subset from the K eigenmode functions; Combining the screened rotor vibration subset and the hull structure vibration subset into a modal matrix, wherein each column of the modal matrix represents an eigenmode function signal; After the modal matrix is ​​mean-centered and whitened, the independent component analysis algorithm is used to separate the modal matrix into rotor vibration signals and hull structure vibration signals.

4. The vibration current characteristic fusion monitoring device for the wear state of the screw pump rotor according to claim 1, characterized in that: The specific steps of identifying the sudden wear point and locating the sudden wear starting point include: Set a delay range. Within the delay range, use the mutual information method to calculate the mutual information value of each wear feature in the wear feature set at different delays, select the target delay, and set the embedding dimension. For the time series of each wear feature in the wear feature set, phase space reconstruction is performed; For each wear feature phase space, the neighborhood radius is set according to the statistical value of the phase space vector distance. For each pair of phase space vectors in the phase space, the distance between the phase space vectors is calculated to calculate the correlation integral of the wear feature. The correlation integral is logarithmically transformed and linearly fitted to obtain the slope of the fitting line as the correlation dimension. The wear feature importance scores of the wear feature set are weighted and normalized to assign weights to each wear feature to obtain the fusion correlation dimension.

5. The vibration current characteristic fusion monitoring device for the wear state of the screw pump rotor according to claim 4, characterized in that: The specific steps of identifying the sudden wear point and locating the sudden wear starting point also include: Based on the historical data of the screw pump rotor, a correlation dimension threshold is configured. If the fused correlation dimension is greater than the correlation dimension threshold, it is determined that a sudden wear point exists and is marked. Set a backtracking cycle to form a backtracking window. Within the backtracking window, perform a backtracking search starting from the sudden wear point. Calculate the relative change rate of the fusion correlation dimension at each time point to calculate the path cost at each time point within the backtracking window and obtain the path cost curve within the backtracking window. Calculate the slope of the path cost curve at each time point in the backtracking window, filter the slope mutation points in the backtracking window, and sort them in chronological order. The time point where the slope mutation point is far away from the sudden wear point is taken as the starting point of the sudden wear.

6. The vibration current feature fusion monitoring device for the wear state of a screw pump rotor according to claim 1, characterized in that: The specific steps of constructing the multivariate time series matrix include: Obtain the speed pulse signal of the motor shaft as the physical synchronization reference for the rotor vibration signal and current signal, generate a timestamp mark, and establish a synchronous clock reference coordinate system; Performing time stamps on the rotor vibration signal and the current signal, and mapping the time stamps of the rotor vibration signal and the current signal to a synchronous clock reference coordinate system; Determine whether to trigger interpolation correction based on the accumulated timestamp deviation between the rotor vibration signal and the current signal; If the interpolation correction is triggered, the following steps are performed: the curvature value of each sampling point of the rotor vibration signal and the current signal is calculated, the local curvature extreme value points are selected as interpolation candidate nodes, and an interpolation node sequence is selected from the interpolation candidate nodes; Based on the interpolation node sequence, the corresponding timestamps of the rotor vibration signal and the current signal are obtained respectively, and the time offset between the interpolation nodes at the same position of the rotor vibration signal and the current signal is calculated; The calculated time offset is used to construct a time deviation matrix to obtain the nonlinear mapping relationship between the rotor vibration signal and the current signal; Based on the current signal time series, the interpolation time points of the rotor vibration signal are obtained through nonlinear mapping, and the aligned rotor vibration signal is generated using the interpolation algorithm. Based on the rotor vibration signal time series, the interpolation time points of the current signal are obtained through nonlinear mapping, and the interpolation algorithm is used to generate the aligned current signal. The aligned rotor vibration signals and current signals are arranged in time sequence to generate a spatiotemporally aligned multivariate time series matrix.

7. The vibration current feature fusion monitoring device for the wear state of a screw pump rotor according to claim 1, characterized in that: The specific steps of dividing the maintenance levels include: When a sudden wear point is identified, the sudden wear probability of the screw pump rotor at the sudden wear point is obtained, and the wear degree per unit time is obtained based on the historical wear data of the screw pump rotor during normal operation; Set a correction factor to correct the wear rate per unit time according to the probability of sudden wear to obtain the corrected wear rate; Determine the wear limit of the screw pump rotor when it reaches scrapping or requires major repairs, obtain the cumulative usage time of the screw pump rotor, the number of sudden wear points, and the probability of sudden wear of the screw pump rotor at the sudden wear points, so as to obtain the current wear amount of the screw pump rotor and use the corrected wear rate to predict the remaining life; Maintenance thresholds are set, including an upper maintenance threshold and a lower maintenance threshold, and the screw pump rotor is divided into different maintenance levels according to the predicted remaining life.

Images