Method for predicting residual life of typical electromechanical equipment of unmanned ship

Abstract

The invention discloses a method for predicting the residual life of typical electromechanical equipment of an unmanned ship, and the method comprises the following steps: 1), collecting the vibration, acceleration, stress and temperature data of the equipment, and carrying out the preprocessing; 2) carrying out signal feature extraction, and obtaining feature parameters capable of reflecting the equipment state from the signals; 3) constructing an equipment health degree curve for the single signal, obtaining a state evaluation result of the single signal by adopting a state evaluation method based on mahalanobis distance measurement, and fusing to obtain an equipment health state; and 4) establishing a residual life prediction model between the equipment health degree and the accumulated operation time, and performing equipment state estimation and residual life prediction according to the current operation monitoring data of the electromechanical equipment. According to the method, the mutual influence between different devices is considered, the prediction model is established by collecting the device operation data to obtain the remaining life of the typical electromechanical device of the unmanned ship, and the method has higher reliability and objectivity.

Description

Technical Field

[0001] The invention relates to ship equipment maintenance technology, and in particular to a method for predicting the remaining life of typical electromechanical equipment of an unmanned boat. Background Art

[0002] With the rapid development of intelligent technology, unmanned platforms are rapidly applied to various fields. Electromechanical equipment is one of the core system equipment in unmanned platforms. However, when the electromechanical equipment of unmanned platforms fails or degrades, it may cause the function of the electromechanical system to be reduced or lost, affecting the mission of the entire equipment, and even causing catastrophic economic losses and safety accidents. Therefore, it is particularly important to establish a remaining life prediction system for electromechanical equipment.

[0003] Traditional equipment remaining life prediction methods require human participation and rely heavily on expert knowledge and experience, which conflicts with the requirement to reduce human participation in unmanned platforms. However, unmanned platforms are equipped with a large number of sensors, which provide a large amount of equipment operation data. How to effectively predict the remaining life of unmanned platform electromechanical equipment based on massive data is of great practical significance.

[0004] The data-driven remaining life prediction method for typical electromechanical equipment of unmanned boats is based on statistical knowledge and machine learning technology. It does not require knowledge of the physical degradation mechanism of the equipment. It only needs to start from the signals collected by the sensors, through data processing and analysis and modeling of the data itself, to realize the evaluation of the remaining life of the equipment. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide a method for predicting the remaining life of typical electromechanical equipment of an unmanned boat in view of the defects in the prior art.

[0006] The technical solution adopted by the present invention to solve the technical problem is: a method for predicting the remaining life of typical electromechanical equipment of an unmanned boat, comprising the following steps:

[0007] 1) Collect the vibration, acceleration, stress and temperature data of the equipment and perform preprocessing;

[0008] 2) Extract signal features and obtain characteristic parameters that can reflect the status of the device from the signal;

[0009] Among them, feature extraction includes extraction of time domain features and time-frequency domain features;

[0010] The extracted time domain signal has mean feature, root mean square value feature, variance feature, kurtosis feature, and waveform factor feature;

[0011] Frequency domain features are to decompose the original signal into sine waves of different frequencies through Fourier transform of the time domain signal, extract the amplitude and phase information of the signal in each frequency band, and combine it with the time domain to form time-frequency domain features;

[0012] 3) Construct the equipment health curve for a single signal, use the state assessment method based on Mahalanobis distance measurement to obtain the state assessment result of a single signal, and fuse it to obtain the equipment health status;

[0013] The larger the Mahalanobis distance, the farther the feature point to be evaluated is from the healthy baseline, and the more serious the performance degradation is; the smaller the Mahalanobis distance, the closer the feature point to be evaluated is to the healthy baseline, and the better the performance level is;

[0014] 4) Establish a remaining life prediction model between equipment health and its cumulative operating time, and estimate the equipment status and predict the remaining life based on the current operating monitoring data of the electromechanical equipment.

[0015] According to the above scheme, in step 3), the importance of the equipment is weighted according to each signal indicator, and then the health status of the equipment is fused.

[0016] According to the above scheme, in step 3), the importance of the equipment is weighted according to each signal indicator, and then the equipment health status is integrated. The equipment importance weight includes the expert weight and the equipment objective weight. The coefficient of variation is introduced, and the weighted sum of the expert weight and the equipment objective weight is maximized to obtain the comprehensive weighting of each signal indicator based on the coefficient of variation.

[0017] According to the above scheme, in step 3), when evaluating the state of a single signal, the signal parameters are sliced, feature points are constructed through feature extraction, and the Mahalanobis distance between the feature points and the healthy baseline is calculated.

[0018] According to the above scheme, in step 3), the Mahalanobis distance calculation process is as follows:

[0019] Calculate for each signal vector x i The average value of:

[0020]

[0021] Where j is the slice number where the signal is located, n is the number of slices; x ij The value of the signal vector in slice j;

[0022] Calculate for each vector x i The standard deviation of i :

[0023]

[0024] Orthogonalize the eigenvectors to obtain Z ij , and transpose it to

[0025]

[0026] Calculate the correlation matrix A of the orthogonal matrix:

[0027]

[0028] The Mahalanobis distance is:

[0029]

[0030] According to the above scheme, in step 4), a remaining life prediction model is established based on the BP neural network to perform equipment state estimation and remaining life prediction.

[0031] The beneficial effects produced by the present invention are:

[0032] 1. The method of the present invention takes into account the mutual influence between different equipment, and establishes a prediction model by collecting equipment operation data to obtain the remaining life of typical electromechanical equipment of unmanned boats, which has higher reliability and objectivity.

[0033] 2. Since the offshore unmanned platform equipment operates in harsh environments and lacks real-time maintenance by personnel, the remaining life prediction and evaluation method of its electromechanical equipment can effectively predict the safe status of equipment operation and guide the maintenance of electromechanical equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] The present invention will be further described below with reference to the accompanying drawings and embodiments, in which:

[0035] Figure 1 is a method flow chart of an embodiment of the present invention;

[0036] Figure 2 It is a schematic diagram of vitality indicators of the water supply pump of an unmanned boat according to an embodiment of the present invention. DETAILED DESCRIPTION

[0037] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0038] like Figure 1 As shown, a method for predicting the remaining life of typical electromechanical equipment of an unmanned boat includes the following steps:

[0039] 1) Collect equipment vibration, acceleration, stress, and temperature data and perform preprocessing. Preprocessing uses data cleaning to improve data quality by filling missing values, smoothing noise data, and removing outliers.

[0040] 2) Extract signal features and obtain characteristic parameters that can reflect the status of the device from the signal;

[0041] Among them, feature extraction includes extraction of time domain features and time-frequency domain features;

[0042] The extracted time domain signal has mean feature, root mean square value feature, variance feature, kurtosis feature, and waveform factor feature;

[0043] Frequency domain features are to decompose the original signal into sine waves of different frequencies through Fourier transform of the time domain signal, extract the amplitude and phase information of the signal in each frequency band, and combine it with the time domain to form time-frequency domain features;

[0044] The time-frequency domain features are joint features of the time domain and frequency domain, which are used to reflect the changes of the signal spectrum over time;

[0045] 3) Construct the equipment health curve for a single signal, use the state assessment method based on Mahalanobis distance measurement to obtain the state assessment result of a single signal, and fuse it to obtain the equipment health status;

[0046] When evaluating the status of a single signal, the signal parameters are sliced, feature points are constructed through feature extraction, and the Mahalanobis distance between the feature points and the healthy baseline is calculated;

[0047] The Mahalanobis distance calculation process is as follows:

[0048] Calculate for each signal vector x i The average value of:

[0049]

[0050] Where j is the slice number where the signal is located, n is the number of slices; x ij The value of the signal vector in slice j;

[0051] Calculate for each vector x i The standard deviation of i :

[0052]

[0053] Orthogonalize the eigenvectors to obtain Z ij , and transpose it to

[0054]

[0055] Calculate the correlation matrix A of the orthogonal matrix:

[0056]

[0057] The Mahalanobis distance is:

[0058]

[0059] The larger the Mahalanobis distance is, the farther the feature point to be evaluated is from the healthy baseline, and the performance degradation is more serious; the smaller the Mahalanobis distance is, the closer the feature point to be evaluated is to the healthy baseline, and the performance is better;

[0060] 4) Establish a remaining life prediction model between the health of the equipment and its accumulated operating time, and estimate the equipment status and predict the remaining life based on the current operating monitoring data of the electromechanical equipment;

[0061] Remaining life prediction model based on BP neural network;

[0062]

[0063] y=f(net j )

[0064] Among them, x1, x2, ..., x t is the extracted eigenvalue, as input; ω ij is the connection weight; f() is the activation function of the neural network; θ j is the bias value of the neural network, and y is the output value of the neuron.

[0065] In this embodiment, nonlinear Sigmoid function and Logistics function are used as activation functions. Sigmoid function is also called S-type function, and its value range is (0, 1); Logistics function is also called hyperbolic tangent function, and its value range is (-1, 1).

[0066] Taking the long-term operating status monitoring data of an unmanned ship's water supply pump as an example, the total data length is 3000h; the pump product status related parameters include the X-axis vibration of the motor and pump connection body, the Y-axis vibration of the motor and pump connection body, etc., a total of 15 parameters.

[0067] 1. First, preprocess the signal and extract its features to obtain characteristic parameters that can reflect the status of the equipment.

[0068] The specific process of this step is as follows: extract statistical parameters from the signal to describe the relationship between the physical signal and time, and reflect the trend of the physical signal changing with time; through Fourier transform, decompose the original signal into sine waves of different frequencies to obtain the amplitude and phase information of the signal in each frequency band. Finally, combine the time domain and frequency domain to form the time-frequency domain features, reflecting the change of the signal spectrum over time.

[0069] For a given type of water supply pump monitoring data, 10 time domain features are extracted to form its feature space. The 10 signal data are subjected to Fourier transformation to generate frequency domain amplitude and phase information, which are combined with the time domain to form a time-frequency domain feature map.

[0070] 2. Use the distance measurement criterion to judge the status of the equipment, obtain the status evaluation result of a single signal and the equipment health curve constructed using a single signal.

[0071] When conducting health assessment, the parameters are sliced, feature points are constructed through feature extraction, and the Mahalanobis distance between the feature points and the health baseline is calculated. The larger the Mahalanobis distance, the farther the feature point to be evaluated is from the health baseline, and the more serious the performance degradation; the smaller the Mahalanobis distance, the closer the feature point to be evaluated is from the health baseline, and the better the performance.

[0072] 3. Establish a remaining life prediction model between equipment health and its cumulative operating time, and estimate the equipment status and predict the remaining life based on the equipment's current operating monitoring data.

[0073] The following remaining life prediction models can be used: a performance degradation prediction model based on a BP neural network model, a performance degradation prediction model based on a long short-term memory neural network, and a performance degradation prediction model based on a random forest.

[0074] 1) Performance degradation prediction model based on BP neural network model

[0075] BP neural network is composed of artificial neurons of artificial neural network as the basic unit. The basic principle of the network is the basic principle of iterative operation between neurons. For n initial input signal values ​​x1, x2, ..., x n is the extracted feature value, which is equivalent to the output value of 1, 2, ..., n neurons in the previous layer of back propagation; ω i1 ,ω i2 ,……,ω ij is the connection weight between the previous layer 1, 2, ..., n neurons and the next layer neurons, and the initial value is randomly generated by the system; f() is a nonlinear function, which serves as an excitation function of the neural network; θ j is the bias value of the neural network, and the initial value is randomly generated by the system; y is the output value of the neuron. The specific action process is as follows:

[0076]

[0077] y=f(net j )

[0078] Usually, BP neural network uses nonlinear Sigmoid function and Logistics function as activation function. Sigmoid function is also called S-type function, and its value range is (0,1); Logistics function is also called hyperbolic tangent function, and its value range is (-1,1).

[0079] 2) Performance degradation prediction model based on long short-term memory neural network

[0080] Long short-term memory (LSTM) neural network is developed on the basis of recursive neural network (RNN), and it has been widely used in many fields. LSTM has a three-layer network structure of input layer, hidden layer and output layer. The uniqueness of LSTM is that multiple gate-controlled memory modules are set in the hidden layer. Compared with recursive neural network, LSTM can effectively overcome the unit gradient disappearance phenomenon during operation. LSTM has multiple gate structures such as input gate, forget gate, output gate, etc.

[0081] 3) Performance degradation prediction model based on random forest

[0082] A random forest is a model made up of many decision trees. Each tree is trained on a random sample of data points. These samples are drawn repeatedly (called bootstrapping), which means that some samples are used multiple times to train a single tree (this can be disabled if desired). A random forest combines hundreds or thousands of decision trees, training each one on a slightly different set of observations (data points are drawn repeatedly), and splitting the nodes in each tree based on a limited number of features. The final prediction of a random forest is the average of the predictions of each individual tree.

[0083] Using the characteristics of a certain type of unmanned boat water pump equipment in normal state as training samples, the degradation data of the water pump is evaluated. The remaining life of the water pump is quantitatively evaluated with a vitality index between 0 and 1. The larger the distance, the greater the difference between the water pump and the normal state, and the less the remaining life; conversely, the smaller the difference from the normal state, the more remaining life. The vitality characteristic index of the water pump is as follows: Figure 2 shown.

[0084] It should be understood that those skilled in the art can make improvements or changes based on the above description, and all these improvements and changes should fall within the scope of protection of the appended claims of the present invention.

Claims

1. A method for predicting the remaining life of typical electromechanical equipment of unmanned boats, characterized in that: The following steps are involved: 1) Collect the vibration, acceleration, stress and temperature data of the equipment and perform preprocessing; 2) Extract signal features and obtain characteristic parameters that can reflect the status of the device from the signal; Among them, feature extraction includes extraction of time domain features and time-frequency domain features; The extracted time domain signal has mean feature, root mean square value feature, variance feature, kurtosis feature, and waveform factor feature; Frequency domain features are to decompose the original signal into sine waves of different frequencies through Fourier transform of the time domain signal, extract the amplitude and phase information of the signal in each frequency band, and combine it with the time domain to form time-frequency domain features; 3) Construct the equipment health curve for a single signal, use the state assessment method based on Mahalanobis distance measurement to obtain the state assessment result of a single signal, and fuse it to obtain the equipment health status; The larger the Mahalanobis distance, the farther the feature point to be evaluated is from the healthy baseline, and the more serious the performance degradation is; the smaller the Mahalanobis distance, the closer the feature point to be evaluated is to the healthy baseline, and the better the performance level is; 4) Establish a remaining life prediction model between equipment health and its cumulative operating time, and estimate the equipment status and predict the remaining life based on the current operating monitoring data of the electromechanical equipment.

2. The method for predicting the remaining life of typical electromechanical equipment of unmanned boats according to claim 1 is characterized in that: In the step 3), when the state evaluation of a single signal is performed, the signal parameters are sliced, feature points are constructed through feature extraction, and the Mahalanobis distance between the feature points and the healthy baseline is calculated.

3. The method for predicting the remaining life of typical electromechanical equipment of unmanned boats according to claim 1 is characterized in that: In the step 3), the Mahalanobis distance calculation process is as follows: Calculate for each signal vector x i The average value of: Where j is the slice number where the signal is located, n is the number of slices; x ij The value of the signal vector in slice j; Calculate for each vector x i The standard deviation of i : Orthogonalize the eigenvectors to obtain Z ij , and transpose it to Calculate the correlation matrix A of the orthogonal matrix: The Mahalanobis distance is:

4. The method for predicting the remaining life of typical electromechanical equipment of unmanned boats according to claim 1 is characterized in that: In the step 3), the importance of the equipment is weighted according to each signal indicator, and then the health status of the equipment is integrated.

5. The method for predicting the remaining life of typical electromechanical equipment of unmanned boats according to claim 1 is characterized in that: In the step 3), the importance of the equipment is weighted according to each signal indicator, and then the equipment health status is integrated. The equipment importance weight includes the expert weight and the equipment objective weight. The coefficient of variation is introduced, and the weighted sum of the expert weight and the equipment objective weight is maximized to obtain the comprehensive weighting of each signal indicator based on the coefficient of variation.

6. The method for predicting the remaining life of typical electromechanical equipment of an unmanned boat according to claim 1 is characterized in that: In the step 4), a remaining life prediction model is established based on the BP neural network to perform equipment state estimation and remaining life prediction.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images