Ship diesel engine vibration fault early warning method based on improved VMD

Abstract

The application discloses a ship diesel engine vibration fault early warning method based on improved VMD, wherein a super parameter optimization is performed by combining a frost ice algorithm RIME and an improved variational mode decomposition VMD, a VMD signal decomposition is performed on an optimal combination to obtain a plurality of IMF values to reconstruct vibration signal sequence data, time domain and frequency domain feature extraction is performed on the reconstructed vibration signal sequence data to obtain time domain feature parameters and frequency domain feature parameters, a Pearson coefficient between each feature parameter in a mixed domain feature set and a warning value is calculated and obtained, a feature parameter satisfying a preset Pearson coefficient threshold is selected as an optimized feature parameter, a residual error between a predicted root mean square value and an actual root mean square value is obtained based on an optimal BiGRU algorithm model to obtain a residual error threshold range of a fault-free operation of the ship diesel engine, a RIME-VMD-BIGRU model for ship diesel engine vibration fault early warning is built, automatic prediction of the ship diesel engine vibration fault is realized, crew members can be reminded of possible faults of the ship, dependence on expert experience is reduced, and the model has higher diagnostic accuracy compared with other early warning algorithms.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of ship diesel engine fault diagnosis, and particularly relates to a ship diesel engine vibration fault early warning method based on improved VMD. BACKGROUND

[0002] As a kind of reciprocating internal combustion engine, the ship diesel engine is widely used in ship power devices and is an important output source of ship power. However, due to the complex working environment and long-term use, the ship diesel engine often fails, and due to its compact and complex structure, some moving parts are usually accompanied by large mechanical load and thermal load, and the inherent characteristics of variable working conditions, which easily lead to complex cause and effect relationship of failure symptoms. At present, the methods for predicting the failure of the ship diesel engine mainly include the following:

[0003] 1. Fault diagnosis based on physical model: the fault prediction and early warning method based on physical model often needs researchers to build accurate mathematical or physical models to describe the working process of the research object. By analyzing the difference between the actual behavior of the system and the expected behavior of the model, it is detected whether the system is in an abnormal state, the possible faults of the system are located, and the relevant management personnel are warned. Although the fault prediction and early warning method based on physical model can accurately realize the fault prediction and early warning of the equipment, the prerequisite is to establish a very accurate mathematical or physical model of the monitored research object. Since the structure of the marine diesel engine system is very complex, the working environment is very harsh, and the working condition is variable, it is difficult to establish a mathematical model of the marine diesel engine. Therefore, the fault prediction and early warning method based on physical model is obviously not suitable for the fault early warning research of the marine diesel engine.

[0004] 2. Fault prediction and early warning method based on empirical model: the fault diagnosis method based on empirical model needs to be based on the knowledge and experience of experts when diagnosing the fault state of the equipment, mainly including fault tree-based and expert system-based diagnosis methods. The method based on empirical model indeed relies on the professional knowledge and practical engineering experience of experts, and is influenced by personal experience and preference to some extent. Therefore, if there is a large deviation between the experience of experts and the actual situation, the accuracy of the abnormal state judgment and fault prediction and early warning will be greatly reduced, which may lead to unpredictable consequences. Therefore, the method based on empirical model needs to use the experience of most experts to build an overall expert knowledge base in the field, which can reduce the subjective influence of individuals on the abnormal state through data fusion.

[0005] 3. Reliability model-based method: Reliability models are usually used to analyze past failure history data using mathematical statistics to predict the future possible failures of equipment. These models can help relevant personnel develop effective maintenance plans, thereby reducing the probability of equipment failure and reducing losses. The method based on reliability failure prediction and early warning is essentially to express the time of occurrence of various types of failures or the probability of failure within a certain period of time using a series of different probability density functions through probability and statistics knowledge. The basis is also to analyze the data information collected under different states of the equipment, so some experts and scholars also classify this method into data-driven methods.

[0006] 4. Data-driven method: Data-driven failure prediction and early warning methods have been proposed and widely applied in the industrial field. This method does not need to establish an accurate physical model of the research object, but uses the collected data for various data analysis and artificial intelligence algorithm processing to mine the implicit information in the data and predict the future state of the research object. Compared with model-based and expert knowledge-based failure prediction and early warning techniques, data-driven methods avoid their shortcomings. Therefore, data-driven failure prediction and early warning methods are receiving more and more attention.

[0007] In summary, traditional diesel engine failure warning methods usually rely on expert experience, and maintenance personnel need to passively discover failures based on maintenance experience or existing failure parameter records. However, in the era of unmanned engine rooms and intelligent shipping, the timeliness and accuracy of this prediction method are often insufficient. SUMMARY

[0008] The present application provides a ship diesel engine vibration fault early warning method based on improved VMD to overcome the above technical problems.

[0009] To achieve the above purpose, the technical scheme of the present application is:

[0010] A ship diesel engine vibration fault early warning method based on improved VMD, comprising the following steps:

[0011] S1: Obtain the non-fault history vibration signal sequence data of the ship diesel engine, and perform normalization preprocessing to obtain an optimized data set;

[0012] S2: Based on the RIME frost-ice optimization algorithm, the optimal combination of the decomposition layer number and the penalty factor of the VMD algorithm is obtained based on the optimized data set;

[0013] S3: Based on the optimal combination, the VMD signal decomposition is performed on the optimized data set to obtain a plurality of IMF values to reconstruct the vibration signal sequence data;

[0014] Based on the frequency domain and time domain analysis method, the time domain and frequency domain characteristic parameters of the reconstructed vibration signal sequence data are extracted to obtain the time domain characteristic parameters and the frequency domain characteristic parameters;

[0015] And the hybrid domain feature set is obtained according to the time domain characteristic parameters and the frequency domain characteristic parameters;

[0016] S4: The root mean square value of the original vibration signal sequence data is obtained as a pre-warning value;

[0017] The Pearson coefficient between each characteristic parameter in the hybrid domain feature set and the pre-warning value is calculated and obtained, and the characteristic parameter whose Pearson coefficient meets the preset Pearson coefficient threshold is selected as the optimized characteristic parameter;

[0018] S5: The optimized characteristic parameters are randomly divided into a training set and a test set according to a preset proportion;

[0019] S6: A BiGRU algorithm model is pre-constructed, the optimized characteristic parameters in the training set are taken as feature data, the pre-warning values corresponding to the feature data are taken as label data as inputs of the BiGRU algorithm model, the BiGRU algorithm model is trained based on a loss function to obtain an optimal BiGRU algorithm model, and then an RIME-VMD-BIGRU model for ship diesel engine vibration fault pre-warning is obtained;

[0020] Based on the RIME-VMD-BIGRU model, the residual error between the predicted root mean square value and the actual root mean square value is calculated and obtained according to the test set;

[0021] S7: According to the residual error, a residual error threshold range [A, B] of the ship diesel engine under normal operation is set, A is the minimum residual error value obtained according to the test set, and B is the maximum residual error value obtained according to the test set;

[0022] S8: The current vibration signal sequence data of the ship diesel engine is input into the RIME-VMD-

[0023] BIGRU model, the residual error of the current ship diesel engine is obtained, and it is judged whether the residual error of the current ship diesel engine meets the residual error threshold range [A, B]; if the residual error of the current ship diesel engine meets the residual error threshold range [A, B], it is confirmed that the current ship diesel engine is running normally; otherwise, it is confirmed that the current ship diesel engine is running in failure, and the current ship diesel engine vibration fault pre-warning is output.

[0024] Further, the S2 specifically includes the following steps:

[0025] S21: The maximum number of iterations is set, and the initial frost particle population of the frost ice optimization algorithm RIME is randomly initialized;

[0026] The parameter combination [k, a] of the decomposition layer number and the penalty factor of the VMD algorithm is defined as the frost ice particle;

[0027] S22: According to the VMD algorithm corresponding to each frost ice particle of the initial frost ice particle population, the fault-free historical vibration signal sequence data is iteratively signal decomposed to obtain a plurality of intrinsic mode function IMF values;

[0028] The sample entropy or envelope entropy is taken as the fitness function to determine the fitness value of each intrinsic mode function IMF value, and the minimum fitness value is taken as the current optimal fitness of the optimal frost ice particle;

[0029] S23: According to the current iteration number and the maximum iteration number, the current adhesion coefficient E of the frost ice particle is obtained;

[0030] S24: A first random number r2 and a second random number r3 are defined;

[0031] And determine the size of the first random number r2 and the current adhesion coefficient E;

[0032] If the first random number r2 is greater than or equal to the current adhesion coefficient E, the current iteration number is updated and step S23 is repeatedly executed;

[0033] Otherwise, the frost ice particle position is updated to obtain an updated frost ice particle population, and each frost ice particle in the updated frost ice particle population and its corresponding optimal fitness are obtained;

[0034] According to each frost ice particle in the updated frost ice particle population and its corresponding optimal fitness, the current normalized fitness value of the updated frost ice particle population is confirmed;

[0035] S25: Determine the size of the second random number r3 and the current normalized fitness value;

[0036] If the second random number r3 is greater than or equal to the current normalized fitness value; then update the current iteration number and repeat steps S23 to S24;

[0037] Otherwise, the updated frost ice particle position is updated to obtain the latest frost ice particle population, and the latest optimal frost ice particle in the latest frost ice particle population and its corresponding latest optimal fitness are obtained;

[0038] S26: Determine the size of the latest optimal fitness and the current optimal fitness;

[0039] If the latest optimal fitness is greater than or equal to the current optimal fitness, the current iteration number is updated and steps S23 to S25 are repeatedly executed;

[0040] Otherwise, continue to execute step S27;

[0041] S27: replace the latest optimal fitness with the current optimal fitness, and determine whether the current iteration number reaches the maximum iteration number;

[0042] If not, update the current iteration number and repeat steps S23-S26;

[0043] If yes, the frost and ice particles corresponding to the current optimal fitness are taken as the final frost and ice particles, that is, the parameter combination of the VMD algorithm corresponding to the final frost and ice particles is taken as the optimal combination.

[0044] Further, the time domain feature parameters and the frequency domain feature parameters obtained in S3 are specifically

[0045] The time domain feature parameters include mean value, square root amplitude, root mean square, peak-to-peak value, skewness, kurtosis, peak factor, margin factor, waveform factor, and pulse index.

[0046] The frequency domain feature parameters include average spectral amplitude, variance of spectral amplitude, skewness of spectral amplitude, kurtosis of spectral amplitude, center of gravity frequency, frequency mean square deviation, root mean square frequency, fourth root mean square frequency, ratio of root mean square frequency to square root of spectral amplitude, ratio of frequency mean square deviation to center of gravity frequency, frequency skewness, and frequency kurtosis.

[0047] Further, the loss function for training the BiGRU algorithm model in S6 is

[0048]

[0049] In the formula, n represents the number of characteristic data of the optimized feature parameters; y i represents the true value of the i-th characteristic data; represents the predicted value of the i-th characteristic data.

[0050] The residual error of the predicted root mean square value and the actual root mean square value obtained is expressed as

[0051]

[0052] In the formula, y i represents the true value of the i-th characteristic data; represents the predicted value of the i-th characteristic data.

[0053] Further, the BiGRU algorithm model pre-constructed in S6 includes an input layer, a first GRU unit, a second GRU unit, and a fusion output layer.

[0054] The input layer is used to transmit the input data to the first GRU unit and the second GRU unit, respectively.

[0055] The first GRU unit is used for time forward processing operation on the output of the input layer to obtain forward processing data.

[0056] The fusion output layer is used for merging the forward processing data and the reverse processing data and outputting.

[0057] Compared with the prior art, the present application has the following beneficial effects:

[0058] 1. Strong applicability: the normalization processing of the fault-free historical vibration signal sequence data of the ship diesel engine eliminates the influence of different diagnostic parameters of different dimensions and orders of magnitude on the diagnostic results;

[0059] 2. High efficiency and accuracy: the RIME algorithm is used to optimize and obtain the optimal combination of the decomposition layer number and the penalty factor of the VMD algorithm, avoiding the inaccurate problem of manual selection, fully mining the information in the data, reducing the demand for artificial features, and improving the efficiency and accuracy of the diagnosis;

[0060] 3. Strong adaptability: by introducing the Pearson coefficient, strong correlation features can be effectively selected, the feature dimension is reduced, the waste of computing resources is reduced, and the prediction accuracy and stability are improved;

[0061] 4. Good real-time performance: the online data processing and real-time diagnosis are adopted, the BIGRU is used to accurately predict the root mean square feature and set a threshold line, and the threshold line range is used to timely remind the duty personnel to check the fault, thereby reducing the maintenance cost.

[0062] In summary, the ship diesel engine vibration fault early warning method based on improved VMD can automatically predict the ship diesel engine vibration fault, quickly remind the crew that the ship will have a problem, reduce the dependence on expert experience, and has higher precision early warning compared with the single BiGRU, and can be widely popularized in the field of ship turbine engineering. BRIEF DESCRIPTION OF DRAWINGS

[0063] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or the prior art description will be briefly introduced. 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 creative labor.

[0064] Figure 1 The flowchart of the ship diesel engine vibration fault early warning method based on improved VMD of the present application;

[0065] Figure 2Flow chart of RIME-VMD algorithm in the embodiment;

[0066] Figure 3 Schematic diagram of BiGRU algorithm model in the embodiment;

[0067] Figure 4 Decomposition diagram of simulated vibration signal in the embodiment;

[0068] Figure 5 Comparison diagram of simulated vibration original signal and reconstructed signal in the embodiment;

[0069] Figure 6 Original sequence diagram of ship vibration signal in the embodiment;

[0070] Figure 7 Iterative convergence diagram of envelope entropy ship vibration signal in the embodiment;

[0071] Figure 8 Iterative convergence diagram of sample entropy ship vibration signal in the embodiment;

[0072] Figure 9 IMF frequency spectrum decomposition diagram of envelope entropy ship vibration signal in the embodiment;

[0073] Figure 10 IMF frequency spectrum decomposition diagram of sample entropy ship vibration signal in the embodiment;

[0074] Figure 11 Comparison diagram of original ship vibration signal and reconstructed signal in the embodiment;

[0075] Figure 12 Comparison diagram of VMD decomposition and non-VMD decomposition prediction in the embodiment;

[0076] Figure 13 Prediction result diagram of RIME-VMD-BiGRU model in the embodiment;

[0077] Figure 14 Prediction result diagram of RIME-VMD-GRU model in the embodiment;

[0078] Figure 15 Prediction result diagram of RIME-VMD-LSTM model in the embodiment;

[0079] Figure 16 Prediction result diagram of RIME-VMD-RNN model in the embodiment;

[0080] Figure 17 Residual threshold diagram in the embodiment;

[0081] Figure 18 Fault diagram of residual threshold in the embodiment;

[0082] Figure 19 The core block diagram of the ship diesel engine vibration fault early warning method based on improved VMD in the embodiment. DETAILED DESCRIPTION

[0083] To make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described below in connection with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0084] The embodiment provides a ship diesel engine vibration fault early warning method based on improved VMD, which comprises the following steps: Figure 1 As shown in the figure, the method comprises the following steps: Figure 19 S1: acquiring a non-fault historical vibration signal sequence data of a ship diesel engine and performing normalization preprocessing to acquire an optimized data set;

[0085] S1: acquiring a non-fault historical vibration signal sequence data of a ship diesel engine and performing normalization preprocessing to acquire an optimized data set;

[0086] Specifically, the sample data of the embodiment, i.e., the non-fault historical vibration signal sequence data of the ship diesel engine, comes from a new teaching and training dual-purpose ship "Xin Hongzuan" of Dalian Maritime University. Three groups of vibration original data are collected, the sampling frequency is 4000 Hz, and the test is performed under five load conditions of 25%, 50%, 75%, 85% and 100%, respectively. The test is performed at the following eight measuring points, as shown in Table 1: 1# cylinder head, 8# cylinder head, supercharger, machine body side (between 7\8 cylinders), machine body side (between 1\2 cylinders), base (1 cylinder side), base (8 cylinder side), and generator bearing cover (non-driving end). The embodiment mainly analyzes the data under the 85% load condition. In the first group of data, 20000 data points are recorded at each measuring point, and the total duration is 5 seconds; in the second group and the third group of data, 16000 data points are recorded at each measuring point, and the total duration is 4 seconds;

[0087] In order to eliminate the influence of different diagnostic parameters and orders of magnitude on the diagnostic results, the sample data is normalized, and the normalization formula of the sample data is:

[0088]

[0089] In the formula, X represents the linear normalized data; X represents the original data; Xmax and Xmin represent the maximum value and the minimum value of a certain feature in the original data, respectively; and X represents the normalized data. norm In the formula, X represents the linear normalized data; X represents the original data; Xmax and Xmin represent the maximum value and the minimum value of a certain feature in the original data, respectively; and X represents the normalized data. max In the formula, X represents the linear normalized data; X represents the original data; Xmax and Xmin represent the maximum value and the minimum value of a certain feature in the original data, respectively; and X represents the normalized data. min In the formula, X represents the linear normalized data; X represents the original data; Xmax and Xmin represent the maximum value and the minimum value of a certain feature in the original data, respectively; and X represents the normalized data. In the formula, X represents the linear normalized data; X represents the original data; Xmax and Xmin represent the maximum value and the minimum value of a certain feature in the original data, respectively; and X represents the normalized data.

[0090] Table 1. Data column number and measuring point

[0091] Column number Measurement point 1 1# cylinder head 2 8# cylinder head 3 Supercharger 4 Body side (between 7\8 cylinders) 5 Body side (between 1\2 cylinders) 6 Base (1 cylinder side) 7 Base (8 cylinder side) 8 Generator bearing cover (non-driving end)

[0092] S2: Based on the frost-ice optimization algorithm RIME, the optimal combination of the decomposition layer number and the penalty factor of the VMD algorithm is obtained according to the optimization data set;

[0093] As shown in the specific steps include: Figure 2

[0094] S21: The algorithm parameters of the frost-ice optimization algorithm RIME and the VMD algorithm are initialized, the maximum iteration number H is set, and the initial frost-ice particle population of the frost-ice optimization algorithm RIME is randomly initialized;

[0095] The parameter combination [k, a] of the decomposition layer number and the penalty factor of the VMD algorithm is defined as the frost-ice particle;

[0096] S22: According to the VMD algorithm corresponding to each frost-ice particle in the initial frost-ice particle population, the iteration signal decomposition of the fault-free historical vibration signal sequence data is carried out, and a plurality of intrinsic mode function IMF values are obtained;

[0097] The sample entropy or envelope entropy is taken as the fitness function, the fitness value of each intrinsic mode function IMF value is determined, and the minimum fitness value is taken as the current optimal fitness of the optimal frost-ice particle;

[0098] Among them, the VMD algorithm decomposition process is mainly divided into two parts of construction and solution of variational constraint problem:

[0099] 1) Construction of variational problem: assuming that the original signal is composed of a plurality of band-limited intrinsic mode functions (IMF), the formula for minimizing the sum of the estimated bandwidth of all μ k (t) is constructed, and the solution that minimizes the sum of the estimated bandwidth of all μ k (t) is obtained:

[0100]

[0101] In the formula: is the derivative of time; δ(t) is the impulse signal function; j is the imaginary number; * is the symbol of convolution calculation; ω k is the assumed frequency; f(t) is the original signal;

[0102] 2) Solution of variational problem: after the construction of the variational problem, the quadratic penalty factor a and the Lagrange multiplier λ(t) are introduced to convert each constraint variational problem into an unconstrained variational problem:

[0103]

[0104] ​In addition, the influence of Gaussian noise is greatly reduced by a quadratic penalty term, and the alternating direction method of multipliers (ADMM) of the multiplier is used to update the values of the IMFs alternately after the conversion of the variational problem is completed Until the variational problem is solved, a number of IMF values are obtained.

[0105] S23: Obtain the current adhesion coefficient E of the frost and ice particles according to the current iteration number h and the maximum iteration number H, and the expression of the current adhesion coefficient E is

[0106]

[0107] S24: Define a first random number r2 and a second random number r3.

[0108] Determine the size of the first random number r2 and the current adhesion coefficient E.

[0109] If the first random number r2 is greater than or equal to the current adhesion coefficient E, update the current iteration number and repeat step S23.

[0110] Otherwise, update the position of the frost and ice particles to obtain an updated population of frost and ice particles, and obtain each frost and ice particle in the updated population of frost and ice particles and its corresponding optimal fitness.

[0111] According to each frost and ice particle in the updated population of frost and ice particles and its corresponding optimal fitness, confirm the current normalized fitness value of the updated population of frost and ice particles.

[0112] The method for obtaining the normalized fitness value is

[0113] Define the RIME population of frost and ice particles as R, and the population of frost and ice particles can be represented by x ij , where i represents the serial number of the frost and ice particles in the population of frost and ice particles, and j represents the serial number of the frost and ice particles in the population of frost and ice particles, and the expression is

[0114]

[0115] The fitness value F(S i ) of all frost and ice particles is

[0116]

[0117] According to the fitness value F(S i ) of all frost and ice particles, calculate the normalized fitness value F n (S i ), and the expression is

[0118]

[0119] S25: judging the size of the second random number r3 and the current normalized fitness value;

[0120] If the second random number r3 is greater than or equal to the current normalized fitness value, updating the current iteration number and repeating steps S23 to S24;

[0121] Otherwise, updating the position of the frost ice particle to obtain the latest frost ice particle population, and obtaining the latest optimal frost ice particle in the latest frost ice particle population and the corresponding latest optimal fitness value;

[0122] S26: judging the size of the latest optimal fitness value and the current optimal fitness value;

[0123] If the latest optimal fitness value is greater than or equal to the current optimal fitness value, updating the current iteration number and repeating steps S23 to S25;

[0124] Otherwise, continuing to perform step S27;

[0125] S27: replacing the latest optimal fitness value with the current optimal fitness value, and judging whether the current iteration number reaches the maximum iteration number;

[0126] If not, updating the current iteration number and repeating steps S23 to S26;

[0127] If yes, taking the frost ice particle corresponding to the current optimal fitness value as the final frost ice particle, that is, taking the parameter combination of the VMD algorithm corresponding to the final frost ice particle as the optimal combination.

[0128] The updating formula of the iteration number is h = h + 1;

[0129] The formula for updating the position of the frost ice particle is

[0130] R i,j = R best + r1 cos θ β (δ (Ub ij -Lb ij + Lb ij ), r2 < E

[0131]

[0132] In the formula, R i,j represents the new position of the updated frost ice particle; R best represents the best frost ice particle in the frost ice population; r1 represents a random number in (-1, 1); r1 and cos θ jointly control the moving direction of the particle; β represents an environmental factor; δ represents a random number in (0, 1), which controls the center distance of the frost ice particle; h represents the current iteration number; H represents the maximum iteration number; w represents the number of control step functions; E represents the adhesion coefficient; Ubij Lb ij respectively represent the upper and lower boundaries of the frost ice particle escape space;

[0133] S3: VMD signal decomposition is performed on the optimized data set based on the optimal combination to obtain a plurality of IMF values to reconstruct the vibration signal sequence data;

[0134] Specifically, in the preset Python script embodiment, first, the pd.read-csv command under the pands library is used to read the preprocessed data set csv file into the algorithm, a total of three groups of data, the first group is 20000 groups of 8-dimensional data, the second and third groups are 16000 groups of 8-dimensional data, which are input into the RIME-VMD algorithm, with sample entropy as the fitness function, and the specific parameters are tau: smoothing factor, default 0. DC: whether to process the direct current component, default 0. init: initialization method, default 1. tol: tolerance, default 1e-7. Population size is 10, upper limit of each dimension [4000, 12], lower limit of each dimension [100, 3], and maximum iteration number is 20. Good convergence effect is obtained after three inputs, and the reconstructed data after VMD decomposition is recombined to form a new data set;

[0135] Based on the frequency domain and time domain analysis method, the time domain and frequency domain feature parameters of the reconstructed vibration signal sequence data are extracted, and the time domain and frequency domain feature parameters are obtained;

[0136] And according to the time domain feature parameters and the frequency domain feature parameters, a hybrid domain feature set is obtained, as shown in Table 2;

[0137] The time domain feature parameters include mean, root amplitude, root mean square, peak-to-peak value, skewness, kurtosis, peak factor, margin factor, waveform factor and pulse index;

[0138] The frequency domain feature parameters include average frequency spectrum amplitude, variance of frequency spectrum amplitude, skewness of frequency spectrum amplitude, kurtosis of frequency spectrum amplitude, center of gravity frequency, frequency mean square deviation, root mean square frequency, fourth root mean square frequency, ratio of root mean square frequency to square root of frequency spectrum amplitude, ratio of frequency mean square deviation to center of gravity frequency, frequency skewness and frequency kurtosis.

[0139] Table 2. Ship diesel engine vibration prediction feature parameters

[0140]

[0141] S4: The root mean square value of the original vibration signal sequence data is obtained as a warning value;

[0142] Pearson correlation coefficients between each feature parameter in the mixed domain feature set and the early warning value are calculated and obtained, and a feature parameter with a Pearson correlation coefficient satisfying a preset Pearson correlation coefficient threshold is selected as an optimized feature parameter;

[0143] Specifically, the domain feature parameters in the mixed domain feature set are selected by using the Pearson correlation coefficient. The Pearson correlation coefficient (Pearson Correlation Coefficient) is a statistical quantity for measuring the strength and direction of the linear relationship between two variables. The Pearson correlation coefficient calculation formula is

[0144]

[0145] In the formula, X i ,Y i represent different feature parameter variables in the mixed domain feature set, is the mean value of the feature parameter variable X i ,Y i . The calculated feature Pearson correlation coefficients are shown in Table 3

[0146] Table 3. Feature Pearson correlation coefficients

[0147]

[0148] In this embodiment, the correlation coefficient in the Pearson correlation coefficient is in the range of 0.4-0.6 and has correlation, so the feature parameters with a correlation coefficient greater than 0.5 are selected as the inputs of the BiGRU algorithm, i.e., the root mean square, peak-to-peak value, skewness, kurtosis, peak factor, margin factor, waveform factor, pulse index, average spectral amplitude, and the like are selected as the inputs of the features, so as to complete the feature dimension reduction process, thereby avoiding the long calculation time and the waste of computing resources;

[0149] S5: The optimized feature parameters are randomly divided into a training set and a test set according to a preset proportion;

[0150] S6: A BiGRU algorithm model is pre-constructed, the optimized feature parameters in the training set are taken as feature data, the early warning value corresponding to the feature data is taken as label data, and the BiGRU algorithm model is taken as the input. The BiGRU algorithm model is trained based on a loss function to obtain an optimal BiGRU algorithm model. Then, an RIME-VMD-BIGRU model for ship diesel engine vibration fault early warning is obtained.

[0151] Specifically, as shown in Figure 3 , the pre-constructed BiGRU algorithm model includes an input layer, a first GRU unit, a second GRU unit, and a fusion output layer.

[0152] The input layer is configured to transmit the input data to the first GRU unit and the second GRU unit respectively;

[0153] The first GRU unit is configured to perform a time forward processing operation on the output of the input layer to obtain forward processing data, and the second GRU unit is configured to perform a time reverse processing operation on the output of the input layer to obtain reverse processing data.

[0154] The fusion output layer is configured to merge the forward processing data and the reverse processing data and output.

[0155] The BiGRU algorithm model in this embodiment has stronger memory ability and can efficiently utilize the input information because it can simultaneously consider the influence of the historical state and the future state on the current state. The input feature data is divided into a training set and a test set in a ratio of 0.8:0.2. After multiple manual adjustments, when the number of hidden layer nodes is 128 and the number of layers is 2, the maximum R2 is 0.9956, the minimum MAE and MAPE are 0.0022 and 0.3578% respectively, and the model effect is the best. In addition to the above parameters, the optimizer is Adam, the loss function is MSEloss, the batch size is 100, and the number of training rounds is 100 rounds.

[0156] The loss function for training the BiGRU algorithm model is

[0157]

[0158] In the formula, n represents the number of feature data for optimizing the characteristic parameters; y i represents the true value of the i-th feature data; represents the predicted value of the i-th feature data.

[0159] Based on the RIME-VMD-BIGRU model, the residual error between the predicted root mean square value and the actual root mean square value is calculated and obtained according to the test set.

[0160] The residual error between the predicted root mean square value and the actual root mean square value is obtained, and its expression is

[0161]

[0162] In the formula, y i represents the true value of the i-th feature data; represents the predicted value of the i-th feature data.

[0163] S7: According to the residual error, set the residual error threshold range [A, B] of the fault-free operation of the ship diesel engine, and A is the minimum residual error value obtained according to the test set, and B is the maximum residual error value obtained according to the test set.

[0164] S8: input the acquired current vibration signal sequence data of the ship diesel engine into the RIME-VMD-

[0165] a BIGRU model, acquire a residual of the current ship diesel engine, and determine whether the residual of the current ship diesel engine satisfies a residual threshold range; if the residual of the current ship diesel engine satisfies the residual threshold range, it is determined that the current ship diesel engine is in normal vibration operation; otherwise, it is determined that the current ship diesel engine is in abnormal vibration operation, and a vibration fault early warning of the current ship diesel engine is output.

[0166] The experiment and result analysis of the embodiment are as follows:

[0167] (1) RIME-VMD simulation data result analysis

[0168] After the RIME-VMD model is built, in order to verify the reliability of the model, firstly, a simulation signal is constructed for verification. A three-harmonic composite signal is composed of three pure harmonic signals with different frequencies, and the center frequencies of the component signals of the composite signal are 8 Hz, 68 Hz, and 788 Hz. In order to be more symbolic to the actual signal, Gaussian white noise ζ with a noise intensity of 0.2 is added to the composite signal. The simulation signal calculation formula is

[0169] f(t) = cos(16πt) + 0.5cos(136πt) + 0.05cos(1576πt) + ζ

[0170] In this embodiment, the RIME-VMD is used to decompose the simulation signal, and the composite signal and the IMF decomposition graph thereof are as shown in Figure 4 The simulation data is decomposed by the optimal parameter combination, and is arranged in order from low frequency to high frequency. The RIME can adaptively decompose the simulation signal sequence into 3 subsequences by optimizing the VMD parameters. The volatility of each subsequence is obviously reduced compared with the simulation signal sequence, and no obvious modal aliasing phenomenon occurs, indicating the feasibility of the adaptive decomposition of RIME-VMD. Among the 3 subsequences after decomposition, the volatility of each subsequence is different. The volatility of the subsequence IMF1 is the most gentle, the waveform is the most smooth, and the average amplitude is the largest, reflecting the overall trend. From the subsequence IMF1 to the subsequence IMF3, the volatility and randomness of the subsequences gradually increase, and the average amplitude decreases, reflecting the influence caused by random factors. As shown in Figure 5 the decomposed signal is reconstructed, and compared with the original signal, the noise in the simulation signal is better removed, and the denoised signal is similar to the original composite signal waveform. Therefore, it is proved that the RIME-VMD method proposed in this embodiment is feasible.

[0171] (2) Analysis of actual data results of RIME-VMD model

[0172] Figure 6 For the original sequence of ship vibration signals, the present embodiment uses the Frosting Ice algorithm to optimize the variational mode decomposition, and respectively uses envelope entropy and sample entropy as the fitness function for optimization. The optimization results of the two fitness functions are compared. Using envelope entropy as the fitness function as shown in Figure 7 , the algorithm shows a fast convergence speed. After a few initial iterations, the fitness value quickly stabilizes, indicating that the algorithm quickly finds a local optimal solution. However, due to the large change in fitness value, this may indicate that there is a large volatility and risk of local optimization in the optimization process; in contrast, when using sample entropy as the fitness function as shown in Figure 8 , the algorithm shows continuous improvement in multiple iterations, and although the fitness value changes less, it gradually decreases during the iteration process and finally converges to a relatively stable value. This indicates that the sample entropy fitness function can explore the optimization space more meticulously and has the potential to find a better solution.

[0173] In summary, sample entropy as the fitness function may be more suitable for deep optimization problems, as it can continuously seek further optimization possibilities while maintaining stability. In comparison to the Frosting Ice algorithm optimized variational mode decomposition (VMD) using envelope entropy and sample entropy as the fitness function, sample entropy exhibits obvious advantages in multiple aspects, making its application in signal analysis more attractive. For example, Figure 9 The number of IMFs optimized by sample entropy is smaller, usually including only a few main frequency components, which makes the decomposition result more concise. This conciseness not only helps to more easily interpret the main characteristics of the signal, but also avoids the complexity and noise interference that may be introduced due to excessive number of IMFs. In contrast, as shown in Figure 10 , although envelope entropy can provide more detailed decomposition and generate more IMFs, some of the modes may contain overlapping frequency components or noise components, resulting in an unclear decomposition result. Secondly, the frequency spectrum of the IMF optimized by sample entropy shows that the frequency distribution of each IMF is extremely clear and concentrated. This sharp frequency peak indicates that each mode represents an independent frequency component in the frequency domain, greatly improving the accuracy of signal decomposition. This clarity is crucial in frequency analysis, as it allows precise positioning and analysis of specific frequency components in the signal. The IMF frequency range of envelope entropy decomposition is wider, and the frequency component distribution is more dispersed, which may result in more ambiguous frequency bands in the frequency spectrum, reducing the accuracy of frequency analysis.

[0174] In summary, the application of sample entropy in the optimization of VMD decomposition can provide clearer, more concise and easier-to-interpret spectrograms, significantly improving the accuracy of signal decomposition and the operability of practical applications, and thus is superior to envelope entropy in many scenarios.

[0175] Figure 11 The original vibration signal and the reconstructed signal of the ship are compared. As can be seen from the figure, with sample entropy as the fitness function, it can be seen that the noise in the simulation signal is better removed, and the waveform of the denoised signal is similar to that of the original composite signal. The denoising method proposed in this chapter accurately reconstructs the noise-free composite signal, and the best k values output three times are 4.9, 4.5 and 4.83, and the best a values are 4000, 2898.11 and 4000 respectively; because k value is usually an integer, the three k values are all taken as 4 by using the floor method. In order to prove the superiority of RIME-VMD, the ship data optimized by RIME-VMD and the ship data not optimized by RIME-VMD are input into BiGRU algorithm for comparison, as shown in Figure 12;

[0176] Table 4. Comparison of model performance

[0177] Prediction model [R 2 ]]> MAE MAPE RIME-VMD-BiGRU 0.9956 0.0022 0.3578％ BiGRU 0.9924 0.1551 9.3803％

[0178] As can be seen from Table 4, first, from the R 2 value, the R 2 value of the RIME-VMD-BiGRU model is 0.9956, close to 1, indicating that the fitting degree of the model to the data is very high. The R 2 value of the BiGRU model is 0.9924, although it is also relatively high, but slightly lower than RIME-VMD-BiGRU, indicating that its fitting effect is slightly inferior. In terms of MAE, the RIME-VMD-BiGRU model performs more superiorly, with MAE only 0.0022, meaning that the average error between the predicted value and the true value of the model is extremely small, while the MAE of the BiGRU model is 0.1551, which is relatively large, indicating that its prediction accuracy is not as good as that of RIME-VMD-BiGRU. The MAPE index reflects the relative percentage of prediction error. The MAPE of RIME-VMD-BiGRU is only 0.3578%, showing that the error of the model in prediction is very low and almost negligible. In contrast, the MAPE of BiGRU is 9.3803%, which is relatively large, indicating that in practical application, the prediction result of the model may deviate from the actual value more.

[0179] In summary, RIME-VMD-BiGRU is superior to BiGRU in all indicators, showing stronger prediction ability and higher accuracy, and is suitable for use in prediction tasks with high accuracy requirements. According to the evaluation criteria of the three parameters, the data processed by RIME-VMD is more superior to fault prediction.

[0180] To further prove the superiority of RIME-VMD-BiGRU, other algorithms are compared, and ablation experiments are conducted, and the comparison results are as shown in Figures 13 to 16

[0181] Table 5. Comparison of results of different prediction models

[0182] Prediction model [R 2 ]]> MAE MAPE RIME-VMD-BiGRU 0.9956 0.0022 0.3578％ RIME-VMD-GRU 0.9723 0.0056 0.9185％ RIME-VMD-LSTM 0.9697 0.0058 0.9579％ RIME-VMD-RNN 0.9896 0.0037 0.6122％

[0183] As shown in Table 5, the performances of four different prediction models in three key performance indicators (R 2 , MAE, MAPE) are shown, which are RIME-VMD-BiGRU, RIME-VMD-GRU, RIME-VMD-LSTM and RIME-VMD-RNN. Through these indicators, the prediction accuracy and fitting ability of each model can be analyzed in depth.

[0184] First, from the R 2 value, the R 2 values of all models are above 0.96, indicating that their fitting effects on data are very good. Among them, the R 2 value of RIME-VMD-BiGRU model is the highest, reaching 0.9956, showing its strong fitting ability. The R 2 value of RIME-VMD-RNN is 0.9896, followed closely, also showing excellent fitting ability. In comparison, the R 2 values of RIME-VMD-GRU and RIME-VMD-LSTM are 0.9723 and 0.9697 respectively, although slightly lower than the former two, but still at a high level, indicating that their performances in fitting are also very reliable.

[0185] In terms of MAE (Mean Absolute Error), RIME-VMD-BiGRU again shows obvious advantages, with an MAE value of only 0.0022, which means that the average error between its predicted value and actual value is extremely small, and it is the smallest among all models. The MAE of RIME-VMD-RNN is 0.0037, which also performs well. In comparison, the MAE of RIME-VMD-GRU and RIME-VMD-LSTM is 0.0056 and 0.0058 respectively, with slightly larger errors, indicating that their prediction accuracy is slightly inferior.

[0186] ​MAPE (Mean Absolute Percentage Error) is an important indicator to measure the relative percentage of prediction error. In this indicator, RIME-VMD-BiGRU also performs well, with MAPE of only 0.3578%, showing that its prediction error in practical application is very low, almost negligible. The MAPE of RIME-VMD-RNN is 0.6122%, also showing a relatively low error. While the MAPE of RIME-VMD-GRU and RIME-VMD-LSTM is 0.9185% and 0.9579% respectively, the error is relatively large, which shows that in actual prediction, they may have a relatively larger deviation.

[0187] In summary, the RIME-VMD-BiGRU model is superior to other models in various indicators, showing excellent prediction ability and high precision, and is the most ideal prediction model in current research. Although RIME-VMD-RNN is slightly inferior, it still has strong fitting ability and high prediction accuracy. In comparison, RIME-VMD-GRU and RIME-VMD-LSTM also perform well to some extent, but their precision and fitting effect are slightly inferior to the first two. In application scenarios with high prediction accuracy requirements, RIME-VMD-BiGRU is undoubtedly the best choice. According to the evaluation criteria of the three parameters, RIME-VMD-BIGRU is more advantageous than other prediction models.

[0188] (3) Setting of threshold value

[0189] In the application of the prediction model, the embodiment first analyzes the residual error between the model prediction value and the actual value, and sets the maximum and minimum residual error values as the threshold value. As shown in Figure 17 , under normal working conditions, the residual error between the model prediction value and the actual value shows a good distribution state, and all residual errors are kept within the preset upper and lower threshold range. This shows that under the condition of no fault of the system, the prediction accuracy of the model is high, effectively avoiding the phenomenon of false alarm. As can be seen, the set threshold value can effectively constrain the alarm behavior of the system under normal conditions, reducing unnecessary interference.

[0190] In order to further verify the effectiveness of the proposed method, since the data set in this embodiment does not contain fault data, the root mean square value is adjusted lower or higher by using artificial linear adjustment of the data set, so as to simulate the possible fault hidden danger of the diesel engine. As shown in Figure 18As shown in the results, the system successfully triggered the fault alarm when the predicted value deviated significantly from the actual value and exceeded the set upper and lower threshold values. This result strongly demonstrates that the early warning method proposed in this embodiment can timely and effectively detect abnormal conditions during diesel engine operation, especially when the root mean square value experiences abnormal fluctuations. The system can accurately issue a fault warning to prompt the on-duty personnel to perform necessary checks and maintenance. At the same time, this method successfully avoids the possibility of false alarms in normal conditions, ensuring the stability and reliability of the system operation.

[0191] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, and are not limited thereto; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that the technical solutions described in the foregoing embodiments can still be modified, or some or all of the technical features can be replaced by equivalents; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A ship diesel engine vibration fault early warning method based on improved VMD, characterized in that, Specifically comprising the following steps: S1: Obtain the non-fault history vibration signal sequence data of the ship diesel engine, and perform normalization preprocessing to obtain an optimized data set; S2: Based on the frost ice optimization algorithm RIME, the optimal combination of the decomposition layer number and the penalty factor of the VMD algorithm is obtained according to the optimized data set; S3: Based on the optimal combination, the VMD signal decomposition is performed on the optimized data set to obtain a plurality of IMF values to reconstruct the vibration signal sequence data; Based on the frequency domain and time domain analysis method, the time domain and frequency domain feature extraction is performed on the reconstructed vibration signal sequence data to obtain the time domain feature parameters and the frequency domain feature parameters; And according to the time domain feature parameters and the frequency domain feature parameters, a hybrid domain feature set is obtained; S4: Obtain the root mean square value of the original vibration signal sequence data as a warning value; Calculate and obtain the Pearson coefficient between each feature parameter in the hybrid domain feature set and the warning value, and select the feature parameter whose Pearson coefficient meets the preset Pearson coefficient threshold as the optimized feature parameter; S5: The optimized feature parameters are randomly divided into a training set and a test set according to a preset proportion; S6: Pre-construct a BiGRU algorithm model, use the optimized feature parameters in the training set as feature data, use the warning values corresponding to the feature data as label data and as the input of the BiGRU algorithm model, train the BiGRU algorithm model based on the loss function, and obtain the optimal BiGRU algorithm model; and then obtain the RIME-VMD-BIGRU model for ship diesel engine vibration fault warning; Based on the RIME-VMD-BIGRU model, the residual error between the predicted root mean square value and the actual root mean square value is calculated and obtained according to the test set; S7: According to the residual error, set the residual error threshold range [A, B] of the non-fault operation of the ship diesel engine; A is the minimum residual error value obtained according to the test set, and B is the maximum residual error value obtained according to the test set; S8: Input the current vibration signal sequence data of the ship diesel engine into the RIME-VMD-BIGRU model, obtain the residual error of the current ship diesel engine, and judge whether the residual error of the current ship diesel engine meets the residual error threshold range [A, B]; if the residual error of the current ship diesel engine meets the residual error threshold range [A, B], it is confirmed that the current ship diesel engine is running normally; otherwise, it is confirmed that the current ship diesel engine is running in failure, and the current ship diesel engine vibration fault warning is output.

2. The improved VMD-based ship diesel engine vibration fault early warning method according to claim 1, characterized in that, The S2 specifically comprises the following steps: S21: Set the maximum number of iterations, and randomly initialize the initial frost ice particle population of the frost ice optimization algorithm RIME; The parameter combination [k, a] of the decomposition layer number and the penalty factor of the VMD algorithm is defined as the frost ice particle; S22: According to the VMD algorithm corresponding to each frost ice particle of the initial frost ice particle population, the non-fault history vibration signal sequence data is iteratively decomposed to obtain a plurality of intrinsic mode function IMF values; The sample entropy or envelope entropy is used as the fitness function to determine the fitness value of each intrinsic mode function IMF value, and the minimum fitness value is used as the current optimal fitness of the optimal frost ice particle; S23: obtaining a current adhesion coefficient E of the frost and ice particle according to the current iteration number and the maximum iteration number; S24: defining a first random number r2 and a second random number r3; and judging the size of the first random number r2 and the current adhesion coefficient E; if the first random number r2 is greater than or equal to the current adhesion coefficient E, updating the current iteration number and repeating step S23; otherwise, updating the position of the frost and ice particle to obtain an updated frost and ice particle population, obtaining each frost and ice particle in the updated frost and ice particle population and the optimal fitness corresponding thereto; confirming the current normalized fitness value of the updated frost and ice particle population according to each frost and ice particle in the updated frost and ice particle population and the optimal fitness corresponding thereto; S25: judging the size of the second random number r3 and the current normalized fitness value; if the second random number r3 is greater than or equal to the current normalized fitness value, updating the current iteration number and repeating steps S23 to S24; otherwise, updating the position of the updated frost and ice particle to obtain a latest frost and ice particle population, and obtaining a latest optimal frost and ice particle in the latest frost and ice particle population and the latest optimal fitness corresponding thereto; S26: judging the size of the latest optimal fitness and the current optimal fitness; if the latest optimal fitness is greater than or equal to the current optimal fitness, updating the current iteration number and repeating steps S23 to S25; otherwise, continuing to execute step S27; S27: replacing the latest optimal fitness with the current optimal fitness, and judging whether the current iteration number reaches the maximum iteration number; if not, updating the current iteration number and repeating steps S23 to S26; if yes, taking the frost and ice particle corresponding to the current optimal fitness as the final frost and ice particle, that is, taking the parameter combination of the VMD algorithm corresponding to the final frost and ice particle as the optimal combination.

3. The improved VMD-based ship diesel engine vibration fault early warning method according to claim 1, characterized in that, The time domain feature parameters and the frequency domain feature parameters obtained in S3 are specifically The time domain feature parameters include mean value, square root amplitude, root mean square, peak-to-peak value, skewness, kurtosis, peak factor, margin factor, waveform factor, and pulse index; The frequency domain feature parameters include average spectral amplitude, variance of spectral amplitude, skewness of spectral amplitude, kurtosis of spectral amplitude, center of gravity frequency, frequency mean square deviation, root mean square frequency, fourth root mean square frequency, ratio of root mean square frequency to square root of spectral amplitude, ratio of frequency mean square deviation to center of gravity frequency, frequency skewness, and frequency kurtosis.

4. The improved VMD-based ship diesel engine vibration fault early warning method according to claim 1, characterized in that, The loss function for training the BiGRU algorithm model in S6 is In the formula, n represents the number of characteristic data of the optimized characteristic parameter; y i represents the true value of the i-th characteristic data; represents the predicted value of the i-th characteristic data; The residual error between the predicted root mean square value and the actual root mean square value is In the formula: y i represents the true value of the i-th feature data; represents the predicted value of the i-th feature data.

5. The improved VMD-based ship diesel engine vibration fault early warning method according to claim 1, characterized in that, The pre-constructed BiGRU algorithm model in S6 includes an input layer, a first GRU unit, a second GRU unit, and a fusion output layer; The input layer is used to transmit the input data to the first GRU unit and the second GRU unit respectively; The first GRU unit is used to perform time forward processing operation on the output of the input layer to obtain forward processing data; the second GRU unit is used to perform time reverse processing operation on the output of the input layer to obtain reverse processing data; The fusion output layer is used to merge the forward processing data and the reverse processing data and output.

Images