State monitoring signal calibration method and system for marine diesel main engine

Abstract

The invention provides a state monitoring signal calibration method and system for a marine diesel main engine, and relates to the technical field of data processing, and the method comprises the steps: obtaining a state monitoring sensor signal of the marine diesel main engine; extracting time domain features of the state monitoring sensor signals; constructing a time domain feature matrix according to the time domain features of the state monitoring sensor signals; extracting a frequency component of a state monitoring sensor signal through a meta-heuristic optimization algorithm; constructing a frequency domain characteristic matrix according to main frequency components of state monitoring sensor signals; splicing the time domain feature matrix and the frequency domain feature matrix to form a fusion feature matrix; according to the fusion feature matrix, drift signals in the state monitoring sensor signals are estimated through the deep learning technology, the drift signals in the state monitoring sensor signals are removed, and the state monitoring sensor signals are calibrated in real time. According to the invention, the signals of the state monitoring sensor can be calibrated in real time, and the state monitoring accuracy of the marine diesel main engine is improved.

Description

Technical Field

[0001] The invention relates to the technical field of data processing, and in particular to a state monitoring signal calibration method and system for a ship diesel main engine. Background Art

[0002] The ship's diesel main engine is the core device that provides propulsion power for ships and is usually used in large merchant ships, tankers, container ships and other ocean-going vessels. The diesel main engine generates mechanical energy by burning diesel to drive the ship's propulsion system (such as propellers). The condition monitoring of ship diesel main engines is an important means to ensure navigation safety and improve operational efficiency. Condition monitoring refers to the use of sensors and other monitoring equipment to monitor and analyze the key parameters of the diesel main engine (such as temperature, pressure, vibration, current, etc.) in real time to determine its operating status, predict possible failures and take preventive measures.

[0003] In actual environments, condition monitoring sensors work in harsh environments such as high temperature and high vibration of diesel engines for a long time, and are easily affected by interference and drift, resulting in signal errors or deviations. Calibrating the condition monitoring signal of ship diesel engines is the basis for ensuring data accuracy, which can effectively support fault diagnosis, improve operating efficiency, extend equipment life, and meet international standards.

[0004] The commonly used sensor data calibration technology currently uses the Kalman filter method based on recursive estimation theory to estimate and correct the drift in the signal in real time, and then compensates for the drift to achieve calibration of the sensor data.

[0005] However, the sensor calibration method using Kalman filtering has high requirements on the accuracy of the noise model and has limited effect on nonlinear systems. In particular, the noise and drift in the diesel main engine signal have complex dynamic characteristics, which are difficult to be effectively processed by Kalman filtering. It is difficult to obtain accurate and effective status monitoring signals for ship diesel main engines, which affects the accuracy of status monitoring. Summary of the invention

[0006] In order to solve the technical problem that the sensor calibration method using Kalman filtering in the prior art has high requirements on the accuracy of the noise model and has limited effect on nonlinear systems, especially the noise and drift in the diesel main engine signal have complex dynamic characteristics, which are difficult to be effectively processed by Kalman filtering, and it is difficult to obtain accurate and effective state monitoring signals of ship diesel main engines, which affects the accuracy of state monitoring, the present invention provides a state monitoring signal calibration method and system for ship diesel main engines.

[0007] The technical solution provided by the embodiment of the present invention is as follows:

[0008] First aspect:

[0009] An embodiment of the present invention provides a method for calibrating a state monitoring signal of a marine diesel engine, comprising:

[0010] S1: Obtain the status monitoring sensor signal of the ship's diesel main engine;

[0011] S2: performing standardization processing on the state monitoring sensor signal;

[0012] S3: extracting the time domain features of the condition monitoring sensor signal;

[0013] S4: constructing a time domain feature matrix according to the time domain features of the state monitoring sensor signal;

[0014] S5: extracting the frequency component of the condition monitoring sensor signal through a meta-heuristic optimization algorithm;

[0015] S6: constructing a frequency domain feature matrix according to the main frequency components of the state monitoring sensor signal;

[0016] S7: splicing the time domain feature matrix and the frequency domain feature matrix to form a fusion feature matrix;

[0017] S8: According to the fusion feature matrix, through deep learning technology, the drift signal in the state monitoring sensor signal is estimated, the drift signal in the state monitoring sensor signal is removed, and the state monitoring sensor signal is calibrated in real time.

[0018] Second aspect:

[0019] An embodiment of the present invention provides a state monitoring signal calibration system for a marine diesel engine, comprising:

[0020] processor;

[0021] A memory having computer-readable instructions stored thereon, wherein when the computer-readable instructions are executed by the processor, the method for calibrating the state monitoring signal of the ship diesel main engine as described in the first aspect is implemented.

[0022] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0023] In the present invention, the time domain characteristics and frequency domain characteristics of the condition monitoring sensor signal are extracted to fully capture the timing changes and frequency components of the sensor signal. After the time domain and frequency domain feature matrices are spliced ​​into a fusion feature matrix, the deep learning model can simultaneously utilize the time and frequency information to capture the complex dynamic characteristics of noise and drift, estimate the drift signal in the condition monitoring sensor signal, remove the drift signal in the condition monitoring sensor signal, obtain accurate and effective condition monitoring signals of the ship's diesel main engine, perform real-time calibration on the condition monitoring sensor signal, and improve the accuracy of condition monitoring of the ship's diesel main engine. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0025] Figure 1 A schematic flow chart of a method for calibrating a condition monitoring signal of a marine diesel engine provided by an embodiment of the present invention;

[0026] Figure 2 A schematic diagram of the structure of a method for calibrating a condition monitoring signal of a marine diesel engine provided by an embodiment of the present invention;

[0027] Figure 3 A schematic structural diagram of a condition monitoring signal calibration system for a marine diesel main engine provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0028] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0029] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design described as "example" in the present invention should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of the word "example" is intended to present the concept in a specific way. In addition, in the embodiments of the present invention, the meaning expressed by "and / or" can be both, or it can be either of the two.

[0030] In order to make the technical problems, technical solutions and advantages to be solved by the present invention more clear, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0031] Reference Manual Attached Figure 1 , showing a flow chart of a method for calibrating a condition monitoring signal of a marine diesel main engine provided by an embodiment of the present invention.

[0032] Reference Manual Attached Figure 2 , showing a structural schematic diagram of a method for calibrating a condition monitoring signal of a ship diesel main engine provided by an embodiment of the present invention.

[0033] The embodiment of the present invention provides a method for calibrating the state monitoring signal of a ship diesel main engine. The method can be implemented by a device for calibrating the state monitoring signal of a ship diesel main engine. The device for calibrating the state monitoring signal of a ship diesel main engine can be a terminal or a server. The processing flow of the method for calibrating the state monitoring signal of a ship diesel main engine can include the following steps:

[0034] S1: Obtain the status monitoring sensor signal of the ship's diesel main engine.

[0035] Optionally, the state monitoring sensor signal of the ship diesel main engine is specifically: a temperature sensor signal, a pressure sensor signal or a current sensor signal.

[0036] S2: Standardize the condition monitoring sensor signals.

[0037] Optionally, the standardization process may include: filtering process, normalization process, etc.

[0038] S3: Extract time domain features of condition monitoring sensor signals.

[0039] Optionally, the time domain characteristics specifically include: mean, standard deviation, root mean square value, kurtosis, skewness, peak-to-peak value, zero crossing rate, waveform factor, peak factor, pulse factor, margin factor and signal energy.

[0040] S4: Construct a time domain feature matrix based on the time domain characteristics of the condition monitoring sensor signal.

[0041] S5: Extract the frequency components of the condition monitoring sensor signal through a meta-heuristic optimization algorithm.

[0042] It should be noted that the prior art often uses Fourier transform or wavelet transform technology to extract the frequency components of the condition monitoring sensor signal. Fourier transform assumes that the signal is stationary (that is, the frequency characteristics remain unchanged over time). This performs poorly for non-stationary signals with dynamic frequency changes (such as signals of diesel main engines under different working conditions), and it is difficult to accurately describe instantaneous frequency changes. Wavelet transform depends on the selection of the mother wavelet function. Different wavelet functions have different frequency responses to signals. Improper selection may lead to frequency extraction distortion. Although wavelet transform provides time-frequency localization, the resolution depends on the wavelet scale and decomposition level. The resolution adjustment is complex and it is difficult to maintain sufficient accuracy in both high-frequency and low-frequency ranges. In this regard, the present invention adopts a new frequency component extraction method: extracting the frequency components of the condition monitoring sensor signal through a meta-heuristic optimization algorithm.

[0043] Optionally, the meta-heuristic optimization algorithm is specifically: a genetic algorithm, a differential evolution algorithm, a cuckoo search algorithm, a flower pollination optimization algorithm or a whale optimization algorithm.

[0044] Among them, the Flower Pollination Algorithm (FPA) is a meta-heuristic optimization algorithm based on the pollination process of flowers in nature. It searches for the optimal solution by simulating the pollination process of flowers in nature.

[0045] Furthermore, the present invention improves and optimizes the flower pollination optimization algorithm.

[0046] In a possible implementation, S5 specifically includes sub-steps S501 to S508:

[0047] S501: Constructing a fitness function of the flower pollination optimization algorithm with the goal of minimizing the mean square error between the frequency component and the actual signal.

[0048] Specifically, the inverse of the mean square error between the frequency component and the actual signal can be used as the fitness function.

[0049] S502: Initialize the flower individuals and conversion probabilities of the flower pollination optimization algorithm. Each flower individual represents a feasible frequency component combination scheme. Each flower individual is composed of multiple dimensional components, and each component represents a frequency component.

[0050] S503: Generate a random number randomly and determine whether the conversion probability is greater than the random number. If so, proceed to S504 for cross-pollination. Otherwise, proceed to S505 for self-pollination.

[0051] Optionally, the conversion probability is specifically:

[0052]

[0053] Among them, P t represents the transition probability at the tth iteration, P0 represents the basic transition probability, q represents the Weber influence factor, γ represents the scale parameter, m represents the shape parameter, exp represents the exponential function with the natural constant e as the base, and T represents the maximum number of iterations.

[0054] It should be noted that the conversion probability gradually decreases with the increase of the number of iterations, which makes the algorithm have a higher probability of global search in the initial stage, and more inclined to local search at the end of the iteration. Such an adaptive strategy helps the algorithm to conduct extensive exploration in the early stage to fully search for the global optimal solution; and in the later stage, it focuses on local optimization to help quickly converge to the optimal solution.

[0055] S504: When cross-pollination is performed, the individual positions of flowers are updated according to the Levy flight mechanism:

[0056]

[0057] in, represents the position of the i-th flower individual at the t+1-th iteration, represents the position of the i-th flower individual at the t-th iteration, θ represents the step length influencing factor, Levy represents the Levy flight step length, represents the global optimal solution at the tth iteration.

[0058]

[0059] Where Γ represents the standard Gamma function, λ represents the exponential parameter, and s represents the random step variable.

[0060] It should be noted that Lévy flight is a random walk mode with a long tail characteristic, which is suitable for global search. In each iteration, Lévy flight allows individual flowers to make larger position jumps, so that the algorithm can cover a larger search space and avoid falling into local optimality too early. For complex multi-peak optimization problems, the long-distance search characteristics of Lévy flight greatly increase the probability of finding the global optimal solution.

[0061] S505: When self-pollination is performed, the individual positions of the flowers are updated according to the golden sine mechanism:

[0062]

[0063] x1=-π+2π(1-τ)

[0064] x2=-π+2πτ

[0065] in, represents the position of the jth flower individual randomly selected at the tth iteration, represents the position of the kth flower individual randomly selected at the tth iteration, r1 and r2 represent random numbers between 0 and 2π, x1 and x2 represent the self-pollination coefficients, and τ represents the golden section number

[0066] It should be noted that the use of the golden sine mechanism during self-pollination can make full use of the periodic changes of the sine function and the natural balance of the golden section number, so that individual flowers can achieve fine adjustment and direction optimization in local search. This mechanism not only ensures the adequacy of local search, but also avoids the risk of falling into the local optimal solution, thereby improving the convergence efficiency and search quality of the flower pollination optimization algorithm.

[0067] S506: Determine whether the fitness value of the flower individual at the position of the t+1th iteration is greater than the fitness value of the flower individual at the position of the tth iteration. If so, replace the position of the flower individual. Otherwise, the flower individual maintains the current position.

[0068] S507: Update the global optimal solution.

[0069] S508: Determine whether the current number of iterations has reached the maximum number of iterations. If so, output the frequency component combination scheme represented by the flower individual with the highest current fitness. Otherwise, return to continue iteration.

[0070] S6: Construct a frequency domain feature matrix based on the main frequency components of the condition monitoring sensor signal.

[0071] Optionally, multiple key positions of the ship diesel main engine are equipped with condition monitoring sensors. Before S7, the method further includes: extracting spatial features of signals of the condition monitoring sensors at multiple key positions. According to the spatial features of the signals of the condition monitoring sensors, a spatial feature matrix is ​​constructed.

[0072] Among them, key positions are those parts that are prone to change or bear heavy loads during the operation of the diesel engine, which have a greater impact on the health of the engine. Specifically, they can be the combustion chamber, cylinder head, crankcase, turbocharger, cooling system, lubrication system, intake manifold and exhaust manifold.

[0073] Optionally, the spatial feature is specifically: spatial correlation between state monitoring sensor signals at different positions.

[0074] Specifically, the Pearson correlation coefficient can be used to characterize the spatial correlation.

[0075] S7: Concatenate the time domain feature matrix and the frequency domain feature matrix to form a fused feature matrix.

[0076] Optionally, when condition monitoring sensors are configured at multiple key positions of the ship's diesel main engine, S7 specifically includes: splicing the time domain feature matrix, the frequency domain feature matrix and the space feature matrix to form a fused feature matrix.

[0077] It should be noted that the splicing of the time domain feature matrix, frequency domain feature matrix and space feature matrix to form a fusion feature matrix can significantly improve the accuracy and comprehensiveness of the calibration of the ship diesel engine status monitoring signal. The data collected by each sensor at different positions often contain noise, and the noise may be more significant in a single feature. Through multi-dimensional feature fusion, the model can extract the main information in three dimensions: time, frequency and space, thereby weakening the impact of noise on the calibration results.

[0078] S8: According to the fusion feature matrix, through deep learning technology, the drift signal in the condition monitoring sensor signal is estimated, the drift signal in the condition monitoring sensor signal is removed, and the condition monitoring sensor signal is calibrated in real time.

[0079] In a possible implementation, S8 specifically includes sub-steps S801 to S805:

[0080] S801: At the input layer, obtain the fusion feature matrix.

[0081] S802: In the projection layer, the fused feature matrix is ​​projected into a low-dimensional feature space through a global convolution operation.

[0082] Among them, the global convolution operation uses a larger convolution kernel size (such as 11x11 or larger), which can include a larger range of input data in the convolution operation, extract global features in a larger range, and provide the deep learning model with the ability to capture global information. Global convolution is already a common prior art, and the present invention will not be repeated.

[0083] It should be noted that in the projection layer, the high-dimensional fusion feature matrix is ​​projected into the low-dimensional feature space through the global convolution operation, thereby extracting the main information in the signal. The dimensionality reduction operation can reduce noise interference and retain the core features related to drift.

[0084] S803: In the reshaping layer, the projection layer output is reshaped into a tensor of a preset size through an upsampling operation.

[0085] Among them, upsampling is a data reconstruction technology, which is mainly used in deep learning models to increase the resolution of feature maps. Upsampling is widely used in image generation, segmentation tasks and sequence data processing. By restoring low-resolution or low-frequency information to higher-resolution information, the model can restore or process the details of the input data more finely. Upsampling is a common prior art and will not be described in detail in this invention.

[0086] It should be noted that the low-dimensional feature matrix after projection is upsampled in the reshaping layer and reshaped into a tensor of preset size to adapt it to the network structure of the subsequent layer. This step ensures that the structure of the tensor adapts to the input requirements of the subsequent long short-term memory network (LSTM), so that the features can be efficiently input into the subsequent network for time series analysis.

[0087] S804: In the estimation layer, a multi-layer cascaded long short-term memory network structure is used to extract drift features and finally determine the drift estimation value.

[0088] Among them, the Long Short-Term Memory (LSTM) network is a special recurrent neural network (RNN) designed to solve the gradient vanishing and gradient exploding problems faced by standard RNNs on long-term dependency issues. LSTM networks can effectively retain and learn important information in long time series data.

[0089] Specifically, the multi-layer cascaded long short-term memory network structure specifically includes multiple cascaded long short-term memory networks. The hidden layer output of the long short-term memory network of the previous level is used as the input of the long short-term memory network of the next level. Then, the hidden layer output of each level of the long short-term memory network is summarized to obtain a comprehensive hidden state, and the drift estimation value is determined according to the comprehensive hidden state.

[0090] It should be noted that the multi-layer cascade structure can extract different time features in the input data layer by layer. The hidden layer output of each level of LSTM is used as the input of the next level of LSTM, so that the model can gradually capture the deep features in the long time series. The stacking of multiple layers of LSTM can effectively extract the patterns on different time scales in the drift signal, which is especially suitable for the slow drift features that appear in the long-term operation of ship diesel engines, and improves the accuracy of drift estimation.

[0091] For each level of LSTM, hidden state extraction is performed according to the following formula:

[0092] I t =Sigmoid(W XI X t +W HI h t-1 +b I )

[0093] F t =Sigmoid(W XF X t +W HF h t-1 +b F )

[0094] O t =Sigmoid(W XO X t +W HO h t-1 +b O )

[0095] C t '=tanh(W XC X t +W HC h t-1 +b C )

[0096] C t =F t ·C t-1 +I t ·C t '

[0097] h t =O t tanh(C t )

[0098] Among them, I t represents the activation output vector of the input gate at time t, Sigmoid() represents the Sigmoid activation function, and W XI represents the weight matrix between the word segmentation sequence and the input gate, X t Represents the weighted word vector input at time t, W HI represents the weight matrix between the hidden state and the input gate, b I represents the bias term of the input gate, F t represents the activation output vector of the forget gate at time t, W XF Represents the weight matrix between the word segmentation sequence and the forget gate, W HF represents the weight matrix between the hidden state and the forget gate, b F represents the bias term of the forget gate, O t represents the activation output vector of the output gate at time t, W XO Represents the weight matrix between the word segmentation sequence and the output gate, W HO represents the weight matrix between the hidden state and the output gate, b O represents the bias term of the output gate, C t represents the activation output vector of the cell storage unit at time t, C t ' represents the candidate output vector of the cell storage unit at time t, C t-1 represents the activation output vector of the cell storage unit at time t-1, tanh() represents the tanh activation function, W XC Represents the weight matrix between the word segmentation sequence and the cell storage unit, W HC represents the weight matrix between the hidden state and the cell storage unit, b C represents the bias term of the cell storage unit, h t represents the hidden state at time t, h t-1 represents the hidden state at time t-1.

[0099] Furthermore, the hidden layer outputs of each level of LSTM network are summarized according to the following formula to obtain the comprehensive hidden state:

[0100]

[0101] Among them, Ht represents the comprehensive hidden state at time t, Sigmoid() represents the Sigmoid activation function, and W i represents the weight matrix of the i-th level long short-term memory network, h it It represents the hidden state outputted by the hidden layer of the i-th level LSTM network at time t, and n represents the total number of LSTM networks.

[0102] Furthermore, the drift estimation value is determined according to the following formula:

[0103] N t+1 =Sigmoid(W h H t +b h )

[0104] Among them, N t+1 represents the drift estimate at time t+1, Sigmoid() represents the Sigmoid activation function, and W h represents the weight matrix between the hidden layer and the prediction layer, b h Represents the prediction layer bias term.

[0105] It should be noted that the use of a cascaded LSTM network structure in the estimation layer can extract complex drift features from the reshaped tensor. LSTM is good at capturing long-term dependencies and patterns in time series, and can accurately identify drift trends and dynamic changes in signals.

[0106] S805: In the recovery layer, the signal measurement value is subtracted from the drift estimation value to obtain a drift-free state monitoring sensor signal, so as to perform real-time calibration on the state monitoring sensor signal:

[0107] Y t+1 =X t+1 -N t+1

[0108] Among them, T t+1 represents the calibrated state monitoring sensor signal at time t+1, X t+1 The calibrated condition monitoring sensor signal measurement value at time.

[0109] It should be noted that in the recovery layer, the original signal measurement value is subtracted from the drift signal estimated by LSTM to obtain a drift-free calibration signal. The calibrated signal reflects the true measurement value of the sensor and effectively removes the error caused by drift.

[0110] In the present invention, by starting from the multidimensional feature matrix to the deep extraction of the multi-layer cascaded long short-term memory network structure, and then to the real-time calibration of the recovery layer, the entire process can achieve high-precision and real-time signal calibration, ultimately ensuring the accuracy and reliability of the diesel main engine status monitoring data.

[0111] In a possible implementation, when condition monitoring sensors are configured at multiple key positions of a ship's diesel main engine, S8, after S803 and before S804, further includes: in the rearrangement layer, utilizing the spatial correlation of adjacent sensor data, and placing the measurement values ​​of adjacent sensor data in adjacent rows.

[0112] Specifically, the shuffle operation is defined as a matrix transformation. The shuffle matrix is ​​used to shuffle the original sensor data matrix. The shuffle matrix satisfies the following conditions: Each row contains only one 1, indicating that the sensor data at each position will only be mapped to the new position. Each column contains at least one 1, ensuring that all sensor data are included in the shuffled matrix. Through this matrix transformation, a shuffled data matrix can be obtained so that the data of adjacent rows come from spatially adjacent sensor positions.

[0113] Furthermore, in order to ensure that spatially adjacent sensors are arranged in adjacent rows in the rearranged matrix, a distance minimization strategy is used to determine the optimal rearrangement matrix:

[0114]

[0115] Among them, min means minimization, max means maximum value, d means distance, x i represents the i-th row of data, x j represents the jth row of data, d(x i ,x j ) represents the Euclidean distance between the i-th row of data and the j-th row of data.

[0116] Furthermore, the above meta-heuristic optimization algorithm can be used to search and determine the best rearrangement matrix. To avoid repetition, the present invention will not be described in detail.

[0117] It should be noted that after configuring sensors at multiple key locations of a ship's diesel engine, the measurement data of each sensor usually have spatial correlation. For example, adjacent sensors may be subject to similar environmental or mechanical influences. Arranging these related data in adjacent rows can help the model better capture the spatial correlation between different locations.

[0118] In the present invention, the rearrangement layer optimizes the arrangement of the input data, so that the LSTM can more effectively learn and retain spatial correlation information, improve the modeling ability of the spatiotemporal features in the sensor signal, and thus more accurately separate and remove the drift signal.

[0119] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0120] In the present invention, the time domain characteristics and frequency domain characteristics of the condition monitoring sensor signal are extracted to fully capture the timing changes and frequency components of the sensor signal. After the time domain and frequency domain feature matrices are spliced ​​into a fusion feature matrix, the deep learning model can simultaneously utilize the time and frequency information to capture the complex dynamic characteristics of noise and drift, estimate the drift signal in the condition monitoring sensor signal, remove the drift signal in the condition monitoring sensor signal, obtain accurate and effective condition monitoring signals of the ship's diesel main engine, perform real-time calibration on the condition monitoring sensor signal, and improve the accuracy of condition monitoring of the ship's diesel main engine.

[0121] Reference Manual Attached Figure 3 , showing a structural schematic diagram of a condition monitoring signal calibration system for a ship diesel main engine provided by the present invention.

[0122] The present invention also provides a state monitoring signal calibration system 20 for a marine diesel engine, comprising:

[0123] Processor 201;

[0124] The memory 202 stores computer-readable instructions, and when the computer-readable instructions are executed by the processor 201, the state monitoring signal calibration method of the ship diesel main engine as described in the method embodiment is implemented.

[0125] The ship diesel main engine condition monitoring signal calibration system 20 provided by the present invention can execute the above-mentioned ship diesel main engine condition monitoring signal calibration method and achieve the same or similar technical effects. To avoid repetition, the present invention will not go into details.

[0126] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0127] In the present invention, the time domain characteristics and frequency domain characteristics of the condition monitoring sensor signal are extracted to fully capture the timing changes and frequency components of the sensor signal. After the time domain and frequency domain feature matrices are spliced ​​into a fusion feature matrix, the deep learning model can simultaneously utilize the time and frequency information to capture the complex dynamic characteristics of noise and drift, estimate the drift signal in the condition monitoring sensor signal, remove the drift signal in the condition monitoring sensor signal, obtain accurate and effective condition monitoring signals of the ship's diesel main engine, perform real-time calibration on the condition monitoring sensor signal, and improve the accuracy of condition monitoring of the ship's diesel main engine.

[0128] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art who is familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

[0129] There are a few points to note:

[0130] (1) The drawings of the embodiments of the present invention only relate to the structures related to the embodiments of the present invention, and other structures may refer to the general design.

[0131] (2) For the sake of clarity, the thickness of a film or region in the drawings used to describe the embodiments of the present invention is enlarged or reduced, that is, these drawings are not drawn according to the actual scale. It is understood that when an element such as a film, a region or a substrate is referred to as being "on" or "under" another element, the element may be "directly" "on" or "under" the other element or there may be intermediate elements.

[0132] (3) In the absence of conflict, the embodiments of the present invention and the features therein may be combined with each other to obtain new embodiments.

[0133] The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. The protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A method for calibrating a condition monitoring signal of a marine diesel engine, characterized in that: include: S1: Obtain the status monitoring sensor signal of the ship's diesel main engine; S2: performing standardization processing on the state monitoring sensor signal; S3: extracting the time domain features of the condition monitoring sensor signal; S4: constructing a time domain feature matrix according to the time domain features of the state monitoring sensor signal; S5: extracting the frequency component of the condition monitoring sensor signal through a meta-heuristic optimization algorithm; S6: constructing a frequency domain feature matrix according to the main frequency components of the state monitoring sensor signal; S7: splicing the time domain feature matrix and the frequency domain feature matrix to form a fusion feature matrix; S8: According to the fusion feature matrix, through deep learning technology, the drift signal in the state monitoring sensor signal is estimated, the drift signal in the state monitoring sensor signal is removed, and the state monitoring sensor signal is calibrated in real time.

2. The method for calibrating the state monitoring signal of a marine diesel main engine according to claim 1, characterized in that: The state monitoring sensor signal of the ship diesel main engine is specifically: a temperature sensor signal, a pressure sensor signal or a current sensor signal.

3. The method for calibrating the state monitoring signal of a marine diesel main engine according to claim 1, characterized in that: The time domain characteristics specifically include: mean, standard deviation, root mean square value, kurtosis, skewness, peak-to-peak value, zero crossing rate, waveform factor, peak factor, pulse factor, margin factor and signal energy.

4. The method for calibrating the state monitoring signal of a marine diesel engine according to claim 1, characterized in that: The meta-heuristic optimization algorithm is specifically: a genetic algorithm, a differential evolution algorithm, a cuckoo search algorithm, a flower pollination optimization algorithm or a whale optimization algorithm.

5. The method for calibrating the state monitoring signal of a marine diesel main engine according to claim 1, characterized in that: S5 specifically includes: S501: constructing a fitness function of a flower pollination optimization algorithm with the goal of minimizing the mean square error between the frequency component and the actual signal; S502: Initializing flower individuals and conversion probabilities of the flower pollination optimization algorithm, each flower individual represents a feasible frequency component combination scheme, each flower individual is composed of multiple dimensional components, and each component represents a frequency component; S503: randomly generate a random number and determine whether the conversion probability is greater than the random number; if so, proceed to S504 for cross-pollination; otherwise, proceed to S505 for self-pollination; S504: When cross-pollination is performed, the individual positions of flowers are updated according to the Levy flight mechanism; S505: When self-pollination is performed, the individual positions of the flowers are updated according to the golden sine mechanism; S506: Determine whether the fitness value of the flower individual at the position at the t+1th iteration is greater than the fitness value at the position at the tth iteration; if so, replace the position of the flower individual; otherwise, keep the current position of the flower individual; S507: Update the global optimal solution; S508: Determine whether the current number of iterations has reached the maximum number of iterations; if so, output the frequency component combination solution represented by the flower individual with the highest current fitness; otherwise, return to continue iteration.

6. The method for calibrating the state monitoring signal of a marine diesel engine according to claim 1, characterized in that: S8 specifically includes: S801: At the input layer, obtaining the fusion feature matrix; S802: In the projection layer, the fused feature matrix is ​​projected into a low-dimensional feature space through a global convolution operation; S803: In the reshaping layer, reshape the output of the projection layer into a tensor of a preset size through an upsampling operation; S804: In the estimation layer, a multi-layer cascaded long short-term memory network structure is used to extract drift features and finally determine the drift estimation value; S805: In the recovery layer, the signal measurement value is subtracted from the drift estimation value to obtain a drift-free state monitoring sensor signal, so as to perform real-time calibration on the state monitoring sensor signal.

7. The method for calibrating the state monitoring signal of a marine diesel main engine according to claim 6, characterized in that: State monitoring sensors are configured at multiple key locations of the ship's diesel main engine; Before the S7, the method further includes: Extract spatial features of condition monitoring sensor signals at multiple key locations; Constructing a spatial feature matrix according to the spatial features of the state monitoring sensor signal; The S7 is specifically: The time domain feature matrix, the frequency domain feature matrix and the space feature matrix are concatenated to form a fusion feature matrix.

8. The method for calibrating the state monitoring signal of a marine diesel main engine according to claim 7, characterized in that: The spatial feature is specifically: the spatial correlation between the state monitoring sensor signals at different positions.

9. The method for calibrating the state monitoring signal of a marine diesel main engine according to claim 8, characterized in that: In the S8, after the S803 and before the S804, the following further includes: In the rearrangement layer, the spatial correlation of adjacent sensor data is exploited to place the measurements of adjacent sensor data in adjacent rows.

10. A condition monitoring signal calibration system for a marine diesel engine, characterized in that: include: processor; A memory having computer-readable instructions stored thereon, wherein when the computer-readable instructions are executed by the processor, the method for calibrating the condition monitoring signal of a marine diesel main engine according to any one of claims 1 to 9 is implemented.

Images