Marine diesel engine training sample processing method, device and equipment and storage medium

Abstract

The embodiment of the invention discloses a marine diesel engine training sample processing method and device, equipment and a storage medium, and relates to the technical field of sample processing. According to the scheme, the method comprises the steps that frequency components in a second preset frequency range are sequentially removed from each frequency band, a training sample set of the marine diesel engine is established according to all second vibration signals generated after removal processing, and abnormal vibration signals collected from a time domain are expanded into multiple abnormal vibration signals from the frequency domain, so that the abnormal vibration signals of the marine diesel engine are obtained. Therefore, the number of the training samples is increased, the deep learning model can be better trained, and the accuracy of the model in fault recognition of the marine diesel engine is improved.

Description

Technical Field

[0001] The present application relates to the technical field of sample processing, and in particular to a method, device, equipment and storage medium for processing a ship diesel engine training sample. Background Art

[0002] Marine diesel engines work in harsh environments for a long time and are prone to failure. They need to be inspected in a timely manner.

[0003] In the related art, ship diesel engine faults are detected by training deep learning models, but the ship diesel engine fault data is scarce, resulting in poor training results for the deep learning models, which reduces the accuracy of the model in identifying ship diesel engine faults. Summary of the invention

[0004] The embodiments of this specification provide a method for processing ship diesel engine training samples to solve the problem of poor training effect of deep learning models used for ship diesel engine fault detection in the prior art.

[0005] To solve the above technical problems, the embodiments of this specification are implemented as follows:

[0006] In a first aspect, an embodiment of this specification provides a method for processing a ship diesel engine training sample, comprising:

[0007] Acquire a first vibration signal of a collection cycle of a ship diesel engine, wherein the first vibration signal includes an abnormal signal;

[0008] Dividing a first preset frequency range corresponding to the first vibration signal into a plurality of frequency segments; the first preset frequency range does not include a fault frequency;

[0009] Eliminate frequency components within a second preset frequency range in each frequency segment in turn, to obtain a second vibration signal after elimination corresponding to each frequency segment;

[0010] A training sample set of the ship diesel engine is established based on all of the second vibration signals.

[0011] In a second aspect, an embodiment of the present specification provides a processing device for a ship diesel engine training sample, comprising:

[0012] An acquisition module, used for acquiring a first vibration signal of a collection cycle of a ship diesel engine, wherein the first vibration signal includes an abnormal signal;

[0013] A division module, used to divide a first preset frequency range corresponding to the first vibration signal into a plurality of frequency segments; the first preset frequency range does not include a fault frequency;

[0014] A rejection module, configured to sequentially reject frequency components within a second preset frequency range in each of the frequency bands, so as to obtain a second vibration signal after rejection processing corresponding to each of the frequency bands;

[0015] A building module, configured to build a training sample set of the marine diesel engine based on all of the second vibration signals.

[0016] In a third aspect, a processing device for training samples of a marine diesel engine provided by an embodiment of this specification includes a memory, a processor, and a computer program stored on the memory. The processor executes the computer program to implement the method for processing training samples of a marine diesel engine in Solution 1.

[0017] In a fourth aspect, a computer-readable storage medium provided by an embodiment of this specification has a computer program stored thereon. When the computer program is executed by a processor, the method for processing training samples of a marine diesel engine in Solution 1 is implemented.

[0018] An embodiment of this specification achieves the following beneficial effects: sequentially reject frequency components within a second preset frequency range in each of the frequency bands, build a training sample set of the marine diesel engine according to all the second vibration signals generated after rejection processing, expand the abnormal vibration signals collected in the time domain into multiple abnormal vibration signals in the frequency domain, thereby expanding the number of training samples, and further being able to better train a deep learning model and improve the accuracy of the model in identifying faults of the marine diesel engine. Description of the Drawings

[0019] To more clearly illustrate the technical solutions in the embodiments of this specification or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments recorded in this application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0020] Figure 1 It is a schematic flowchart of a method for processing training samples of a marine diesel engine provided by an embodiment of this specification;

[0021] Figure 2 It is a schematic diagram of an application scenario of a method for processing training samples of a marine diesel engine provided by an embodiment of this specification;

[0022] Figure 3 It is a schematic diagram of a normal vibration signal of a marine diesel engine in a normal operating state provided by an embodiment of this specification;

[0023] Figure 4Schematic diagram of abnormal vibration signal under resonance fault condition of marine diesel engine provided by embodiments of this specification;

[0024] Figure 5 Schematic diagram of original abnormal vibration signal and abnormal vibration signal after rejection processing provided by embodiments of this specification;

[0025] Figure 6 Schematic diagram of fault prediction confusion matrix when the number of samples is unbalanced provided by embodiments of this specification;

[0026] Figure 7 Schematic diagram of fault detection confusion matrix when the number of samples is balanced provided by embodiments of this specification;

[0027] Figure 8 Evaluation results of four evaluation indexes under the condition of unbalanced samples for resonance fault diagnosis of marine diesel engine provided by embodiments of this specification;

[0028] Figure 9 Evaluation results of four evaluation indexes under the condition of balanced samples for resonance fault diagnosis of marine diesel engine provided by embodiments of this specification;

[0029] Figure 10 Schematic diagram of the structure of a processing device for marine diesel engine training samples provided by embodiments of this specification;

[0030] Figure 11 Schematic diagram of the structure of a processing device for marine diesel engine training samples provided by embodiments of this specification. Detailed implementation manners

[0031] To make the objectives, technical solutions, and advantages of one or more embodiments of this specification clearer, the technical solutions of one or more embodiments of this specification will be clearly and completely described below in conjunction with the specific embodiments of this specification and the corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this specification, rather than all of the embodiments. Based on the embodiments in this specification, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope protected by one or more embodiments of this specification.

[0032] The following will detail the technical solutions provided by each embodiment of this specification in conjunction with the drawings.

[0033] Specifically describe a processing method for marine diesel engine training samples provided by embodiments of the specification in conjunction with the drawings.

[0034] Figure 1A flowchart of a method for processing a ship diesel engine training sample provided in an embodiment of this specification. From a program perspective, the execution subject of the process can be a program or an application client installed on an application server. On the other hand, from a hardware perspective, the execution subject of the process can be a terminal device, or a detection platform, etc., which is not particularly limited in this embodiment.

[0035] like Figure 1 As shown, the process may include the following steps:

[0036] Step 110: Acquire a first vibration signal of a collection cycle of a ship diesel engine, wherein the first vibration signal includes an abnormal signal.

[0037] In the embodiments of this specification, an acceleration sensor is used to collect vibration signals of a ship diesel engine under operating conditions. The vibration signals within a collection cycle are taken as a group. Abnormal vibration signals under abnormal operating conditions can be collected, and normal vibration signals under normal operating conditions can also be collected. Multiple groups of vibration signals under different working conditions are collected to ensure that the slight changes in the fault mode and the fluctuation characteristics of the normal state can be reflected. The first vibration signal can be an abnormal vibration signal. The abnormal vibration signal can include vibration signals of a ship diesel engine under different fault types, such as a ship diesel engine bearing fault, a ship diesel engine gear fault, and the like.

[0038] Step 120: Divide a first preset frequency range corresponding to the first vibration signal into a plurality of frequency segments; the first preset frequency range does not include a fault frequency.

[0039] In the embodiments of the present specification, the fault frequencies generated when a ship diesel engine fails are all concentrated in the medium and high frequency bands. The first preset frequency range can be the frequency range after removing the medium and high frequency bands. The first preset frequency range corresponding to the frequency domain of the first vibration signal is divided into multiple frequency bands of equal or unequal widths, and the required number of frequency bands are divided according to needs.

[0040] Step 130: Remove frequency components within the second preset frequency range in each frequency segment in turn to obtain a second vibration signal after removal corresponding to each frequency segment.

[0041] In the embodiments of the present specification, the order of eliminating frequency components can be determined according to preset rules. For example, frequency segments are selected in sequence from low frequency to high frequency or from high frequency to low frequency based on the frequency distribution, and the frequency components in the selected frequency segments are eliminated in sequence, and the frequency components within the second preset frequency range in each frequency segment are eliminated respectively.

[0042] For example, for the first time, frequency components within the second preset frequency range are removed from the first frequency segment of the first vibration signal, and other frequency components are retained, so that a new second vibration signal can be obtained based on the other frequency components; for the second time, frequency components within the second preset frequency range are removed from the second frequency segment of the first vibration signal, and other frequency components are retained, obtaining a new second vibration signal. The above removal operation is repeated until each frequency segment has been subjected to a removal operation once, and a new second vibration signal is generated after each removal operation.

[0043] After applying a Boolean mask to each frequency segment, the obtained signals are combined to form the second vibration signal after the removal process. Each second vibration signal is different. These second vibration signals share some basic information (such as frequency segment division, removal rules, etc.), but differ in terms of frequency components, signal strength, signal characteristics, etc.

[0044] The second preset frequency range can be fixed or dynamically adjusted based on the characteristics of the frequency segment.

[0045] Step 140: Based on all of the second vibration signals, establish a training sample set for the marine diesel engine.

[0046] In the embodiments of this specification, a series of second vibration signals are generated according to preset rules and a preset number of times. These abnormal vibration signals are obtained by removing specific frequency components, expanding the number of abnormal vibration signals. All the second vibration signals form a training sample set, which will be used for the training of the subsequent fault diagnosis model.

[0047] It should be understood that the order of some steps in the method described in one or more embodiments of this specification can be interchanged according to actual needs, or some of the steps can also be omitted or deleted.

[0048] In the embodiments of this specification, frequency components within the second preset frequency range are sequentially removed from each of the frequency segments. According to all the second vibration signals generated after the removal process, a training sample set for the marine diesel engine is established. By expanding the abnormal vibration signals collected in the time domain into multiple abnormal vibration signals in the frequency domain, the number of training samples is expanded, and thus the deep learning model can be better trained, improving the accuracy of the model in identifying faults in the marine diesel engine.

[0049] Based on Figure 1 the method in, the embodiments of this specification also provide some specific implementation schemes of this method, which will be described below.

[0050] In the fault diagnosis of marine diesel engines, time-domain and frequency-domain characteristics can reflect the operating state of marine diesel engines from different perspectives. Time-domain characteristics can capture the amplitude, volatility, and statistical properties of signals, while frequency-domain characteristics reveal the frequency components and energy distribution of signals. However, the characteristics in a single domain often struggle to comprehensively describe complex fault patterns, especially resonance faults, whose vibration signals exhibit unique characteristics in both the time domain and the frequency domain.

[0051] The vibration signal of a resonance fault shows periodic changes in amplitude in the time domain and significant enhancement of specific frequency components in the frequency domain. The characteristics in a single domain are difficult to comprehensively capture the multi-dimensional properties of resonance faults, which may lead to misjudgment or missed judgment in the diagnosis results. The working environment of diesel engines is complex and changeable, and the signals may contain a large number of interference components, which also reduces the accuracy of fault identification.

[0052] Optionally, to solve the above problems, in the embodiments of this specification, based on all of the second vibration signals, a training sample set of the marine diesel engine is established, which may specifically include:

[0053] Determine the characteristic parameters of each of the second vibration signals;

[0054] Based on the characteristic parameters, obtain the characteristic vector corresponding to each of the second vibration signals;

[0055] Based on all of the characteristic vectors, establish the training sample set of the marine diesel engine.

[0056] In the embodiments of this specification, the characteristic parameters of each second vibration signal are extracted, and the extracted characteristic parameters are combined into a characteristic vector, which contains multiple dimensions that can describe the characteristics of the vibration signal. The dimension of the characteristic vector depends on the number of extracted characteristic parameters. For example, if 5 time-domain characteristics and 3 frequency-domain characteristics are extracted, then the dimension of the characteristic vector is 8.

[0057] Integrate the characteristic vectors of all the collected second vibration signals to form a training sample set, which can be used for training machine learning or deep learning models to classify, predict, or diagnose the state of marine diesel engines.

[0058] Furthermore, optionally, in the embodiments of this specification, the characteristic parameters include time-domain characteristics and frequency-domain characteristics. Based on the characteristic parameters, to obtain the characteristic vector corresponding to each of the second vibration signals, the method further includes:

[0059] Fuse the time-domain characteristics and the frequency-domain characteristics of each of the second vibration signals respectively to obtain the characteristic vector corresponding to each of the second vibration signals.

[0060] In the embodiments of this specification, for each second vibration signal, its time-domain characteristics and frequency-domain characteristics are analyzed. The time-domain characteristics mainly reflect the variation law of the signal with time, while the frequency-domain characteristics mainly reflect the distribution characteristics of the signal in terms of frequency.

[0061] The time-domain characteristics and frequency-domain characteristics of each second vibration signal are fused to form a comprehensive feature vector, which can more comprehensively reflect the characteristics of the signal, thereby improving the accuracy of fault diagnosis.

[0062] Based on all the feature vectors, a training sample set of the marine diesel engine is established. Each training sample includes a feature vector and a corresponding fault category.

[0063] Specifically, several time-domain characteristics are extracted from each second vibration signal, such as 11 time-domain characteristics. The time-domain characteristics can include mean, standard deviation, root mean square amplitude, RMS root mean square, peak-to-peak value, skewness, kurtosis, peak factor, margin factor, waveform factor, pulse index, and so on.

[0064] Perform a fast Fourier transform on each second vibration signal to extract several frequency-domain characteristics, such as 12 frequency-domain characteristics. The frequency-domain characteristics can include average amplitude, amplitude variance, amplitude skewness, amplitude kurtosis, center frequency, frequency standard deviation, root mean square frequency, kurtosis frequency, shape factor, variance factor, frequency skewness, frequency kurtosis, and so on. Ensure that the extracted characteristics can accurately reflect the fault characteristics of the marine diesel engine, such as imbalance, misalignment, looseness, wear, etc. Table 1 below shows the characteristic factors of the time-domain characteristics and frequency-domain characteristics and the corresponding formulas.

[0065] Table 1

[0066]

[0067]

[0068] In Table 1, the time-domain signal y = [y1, y2, …, y N , N is the length of the signal, and y i is the corresponding i-th sampling point; the frequency-domain signal Y = [Y1, Y2, …, Y K , K is the length of the frequency-domain signal, and Y i is the amplitude of the i-th frequency component, and f i is the corresponding frequency value.

[0069] Time-domain features are extracted from the time variation of signals, which can reflect the amplitude variation and statistical characteristics of signals. However, there are certain limitations. For example, time-domain features cannot directly reflect the frequency components of signals, and there may be obvious differences in frequency between fault features and interference components. Time-domain features are easily affected by noise. Especially in a strong noise environment, fault features may be masked. Frequency-domain features are extracted from the frequency components of signals through Fourier transform, which can reflect the frequency distribution characteristics of signals. But there are also certain limitations. For example, frequency-domain features cannot directly reflect the time variation characteristics of signals, and some fault features may have obvious transient changes in time.

[0070] The fusion of the time-domain features and frequency-domain features of each second vibration signal can be achieved by methods such as simple concatenation, weighted summation, and principal component analysis (PCA). For example, concatenating 11 time-domain features and 12 frequency-domain features to form a 23-dimensional feature vector. Through feature fusion, comprehensively capture the time-frequency domain characteristics of signals, providing richer information for fault diagnosis.

[0071] The fusion of time-domain and frequency-domain features can provide a richer feature set, enhancing the ability to distinguish fault features and interference components. The fused features have stronger robustness to noise and interference, and can extract fault features more accurately.

[0072] In practical applications, in order to better reflect the characteristics of normal vibration signals and keep synchronous processing with abnormal vibration signals, the time-domain features and frequency-domain features of each normal vibration signal can be determined. The time-domain features and frequency-domain features of each normal vibration signal are respectively fused to obtain a feature vector corresponding to each normal vibration signal. Based on this feature vector, a normal training sample set of the marine diesel engine is established.

[0073] Optionally, before the method of sequentially removing the frequency components within the second preset frequency range in each of the frequency segments to obtain the second vibration signal after the removal process corresponding to each of the frequency segments, the method may further include:

[0074] Determine the frequency axis of each of the frequency segments;

[0075] Generate boolean arrays with the same length as each of the frequency axes respectively;

[0076] Generate a boolean mask for each of the boolean arrays based on the second preset frequency range.

[0077] In the embodiments of this specification, for each frequency segment, its frequency axis is determined. The frequency axis represents all possible frequency values within this frequency segment.

[0078] Generate a boolean array with the same length as the frequency axis for each frequency band. Each element of the boolean array corresponds to a frequency point on the frequency axis, and all elements are initially set to true or false.

[0079] Based on the second preset frequency range (i.e., the frequency range to be excluded), generate a boolean mask for each boolean array. The boolean mask is used to identify which frequency components need to be excluded, and its element values correspond to those of the boolean array, but the elements within the second preset frequency range are set to boolean values opposite to the initial values.

[0080] Optionally, in the embodiments of this specification, sequentially excluding the frequency components within the second preset frequency range in each of the frequency bands may specifically include:

[0081] According to the boolean mask, mark the frequency components within the frequency band corresponding to the boolean mask;

[0082] Sequentially set the spectral amplitudes corresponding to the frequency components to zero.

[0083] In the embodiments of this specification, each time the exclusion operation is performed, mark the frequency components to be excluded corresponding to the currently processed frequency band according to the boolean mask,

[0084] Set the spectral amplitudes corresponding to the frequency components marked as to be excluded to zero, and these frequency components will not affect the overall characteristics of the signal.

[0085] Specifically, for a continuous signal x(t), its Fourier transform is:

[0086]

[0087] The signal x(t) can be transformed into a frequency-domain representation through Fourier transform, where f is the frequency, and X(f) represents the complex form of the signal in the frequency domain, including amplitude and phase information.

[0088] After performing the Fourier transform, calculate the frequency resolution. The formula for calculating the frequency resolution is:

[0089]

[0090] where F s is the sampling frequency of the signal, and N is the total number of sampling points of the signal. Then determine the frequency axis, which represents the frequency values corresponding to each frequency-domain component. The formula for the frequency axis is:

[0091]

[0092] Due to the symmetry of the Fourier transform, usually only the positive frequency part is taken, and the final frequency axis can be determined as:

[0093]

[0094] If the frequency range to be removed is [f a , f b , a Boolean array identical to the frequency axis of that frequency segment is created for each frequency segment to mark the frequency components to be removed in each frequency segment. The generation method of the Boolean mask is as follows:

[0095] mask = (frequencies < f a ) ∨ (frequencies > f b )

[0096] The purpose of the spectrum removal operation is to set the signal amplitude of a specific frequency range to zero in the frequency domain, that is, the frequency components within the frequency range [f a , f b will be set to zero, and other frequency components will be retained.

[0097] The Boolean mask in each frequency segment is removed in sequence, and the frequency domain after removal is expressed as:

[0098]

[0099] X′(f) represents the frequency domain representation after removal, that is, within the frequency interval [f a , f b , the spectrum amplitude of the signal is set to zero, and other frequency components remain unchanged. The frequency domain representation X′(f) after removal is converted back to the time-domain signal x′(t) through the inverse Fourier transform:

[0100]

[0101] The obtained x′(t) is the second vibration signal after spectrum removal. Select a first vibration signal (parent sample) x(t). For each first vibration signal, by removing [f a , f b in each frequency segment through spectrum removal, multiple second vibration signals (new samples) are generated. Each new sample returns from the frequency domain to the time domain through the inverse Fourier transform to obtain x′(t).

[0102] In practical applications, the frequency components within the second preset frequency range can also be removed through a band-stop filter. By precisely setting the center frequency and bandwidth of the band-stop filter, the frequency components within the second preset frequency range that are not needed in each frequency segment can be removed.

[0103] Optionally, the method described in the embodiments of this specification may further include:

[0104] Based on the training sample set, train a preset classification model to obtain a trained preset classification model.

[0105] In the embodiments of this specification, the preset classification model may include a support vector machine (SVM), a decision tree, a random forest, a neural network, and so on.

[0106] Based on the normal training sample set established from normal vibration signals and the abnormal training sample set established from abnormal vibration signals, the training sample set is input into the selected preset classification model, and the model parameters are continuously adjusted through an iterative optimization algorithm, thereby constructing a model that can accurately identify new sample categories. By fusing the time-domain features and frequency-domain features of the vibration signal to be detected and inputting them into the trained classification model, it is possible to accurately identify whether the vibration signal is a normal vibration signal or an abnormal vibration signal. If it is an abnormal vibration signal, the corresponding fault category of the abnormal vibration signal can be further identified.

[0107] For ease of understanding, the support vector machine is used as an example for illustration. The support vector machine is a supervised learning algorithm commonly used in classification problems, especially suitable for dealing with binary classification problems. Its core idea is to find an optimal hyperplane to separate data samples of different categories and maximize the interval between these two categories. To perform intelligent diagnosis on the faults of a marine diesel engine, the support vector machine can be used to distinguish the feature vectors of the normal state and the fault state, thereby determining whether the marine diesel engine has a fault.

[0108] Collect the vibration signals of the marine diesel engine, extract the time-domain features and frequency-domain features of the vibration signals, and obtain two-dimensional column feature vectors. These feature vectors are divided into two categories: the two-dimensional column feature vectors in the normal vibration sample Ω n0 where each element is a feature vector, and n1 is the number of normal samples; the two-dimensional column feature vectors in the abnormal vibration sample Ω where each element is a feature vector, and n2 is the number of abnormal samples. Mark the labels of the normal vibration samples as 1 and the labels of the abnormal vibration samples as -1 to obtain the label y fa corresponding to each sample x i i

[0109]

[0110] The goal of the support vector machine is to find an optimal classifier that can maximize the interval between categories (i.e., the widest distance of the decision boundary). The classifier form of the support vector machine is as follows:

[0111]

[0112] where x is the two-dimensional column feature vector to be classified, ω is the optimal column vector to be solved, T is the transpose, and b is the bias of the decision surface.​​​ is a feature mapping function that maps the input feature vector x from the original input space to a higher-dimensional feature space, usually called the kernel mapping. Through the above decision function, the support vector machine can perform classification. When the sample x to be classified u is input into the model, the classification decision rule is as follows:

[0113]

[0114] If D(x u ) > 0, then the sample x u is determined to be a normal vibration signal, that is, the ship diesel engine is operating normally; if D(x u ) ≤ 0, then the sample x u is determined to be an abnormal vibration signal, that is, the ship diesel engine is operating abnormally.

[0115] The support vector machine finds the optimal ω and b by optimizing an objective function. This objective function not only needs to maximize the margin between classes (i.e., minimize the norm of ω), but also needs to tolerate a small number of classification errors (i.e., the existence of slack variables ξ i ). Therefore, the optimization problem of the support vector machine can be expressed as:

[0116]

[0117] where C is a positive scalar called the penalty coefficient, which adjusts the penalty degree of misclassification. ξ i is a slack variable used to measure whether x i is correctly classified; if ξ i > 0, it means that there is a classification error, and the error degree is ξ i . The constraint conditions are:

[0118]

[0119] ξ i ≥ 0, i = 1, 2, …, n1 + n2

[0120] These constraints ensure that most samples are correctly classified while tolerating a small number of classification errors.

[0121] To solve this optimization problem more conveniently, it is usually transformed into the dual space. In the dual space, the optimization problem can be expressed as:

[0122]

[0123] s.t. 0 ≤ Λ ≤ C

[0124] Λ T Y = 0

[0125] where T represents the transpose, is the Lagrange multiplier vector to be adjusted, representing the weight of each sample. 1 and Y are column vectors. For ease of viewing, 1 and Y are transposed into row vectors 1 T and Y T , 1 T =(1, 1, …, 1) represents an (n1 + n2)-dimensional unit vector, represents the label vector. The symbol "≤" represents element-wise comparison. H is a square matrix of dimension (n1 + n2)×(n1 + n2). The elements of H are as follows:

[0126]

[0127] where K(x k , x l ) is the kernel function of x k and x l . It allows for non-linear classification in the feature space, that is, by mapping the samples to a higher-dimensional space to solve the case of linear inseparability in the original input space.

[0128] Through the optimization of the dual problem, the optimal Lagrange multiplier is obtained Then the decision function can be constructed using these parameters. The final decision function can be written as:

[0129]

[0130] where, for i = 1, 2, …, n1 + n2 are the Lagrange multipliers in the optimal solution of the dual problem. K(x i , x) is the value of the kernel function between the sample x i and the sample x to be classified. b * is the optimal bias. In the dual problem, only the samples with non-zero Lagrange multipliers are called support vectors. Support vectors are those samples that affect the decision boundary. Finally, the support vectors, their labels y i , the corresponding Lagrange multipliers and the bias b * together constitute the decision function of the support vector machine.

[0131] Through such a decision-making process, the support vector machine can classify new samples, thus realizing the intelligent diagnosis of ship diesel engine faults.

[0132] Figure 2 is a schematic diagram of the application scenario of a method for processing ship diesel engine training samples provided by an embodiment of this specification.

[0133] As Figure 2As shown, step 210: Collect the normal vibration signals and abnormal vibration signals of the marine diesel engine.

[0134] Using a high-precision vibration velocity sensor, collect the normal vibration signals of the marine diesel engine under normal operating conditions as normal training samples, and collect the abnormal vibration signals of the marine diesel engine under resonance fault conditions as resonance fault training samples.

[0135] Figure 3 This is a schematic diagram of the normal vibration signals of the marine diesel engine under normal operating conditions provided by the embodiments of this specification.

[0136] As Figure 3 shown, collect the vibration velocity waveform of the marine diesel engine in the normal state. Under normal conditions (Normal), collect vibration signals, including amplitude information and time information. The sampling frequency is 4096 Hz, the collection time is 5 seconds, corresponding to 20480 data points. Take every 205 data points as a sample, set the overlap number to 102, and the step size to 103, and a total of 100 normal samples are generated.

[0137] Figure 4 This is a schematic diagram of the abnormal vibration signals of the marine diesel engine under resonance fault conditions provided by the embodiments of this specification.

[0138] As Figure 4 shown, since the resonance fault occurrence frequency of the marine diesel engine is relatively low, under resonance fault conditions (ResonanceFault), collect vibration signals. The collection time is less than 1 s, corresponding to 1132 data points. Keep the length, overlap number, and step size of each sample the same, and a total of 10 fault samples are generated. The ratio of the normal training samples to the resonance fault training samples of the marine diesel engine is 10. The number of normal training samples is much higher than the number of resonance fault training samples, and there is a problem of unbalanced sample numbers. When the ratio of normal training samples to resonance fault training samples is greater than 3, it can be considered that the sample numbers are unbalanced. When the ratio of normal training samples to resonance fault training samples is 1, it can be considered that the sample numbers are balanced.

[0139] Step 220: For the abnormal vibration signals, according to the Boolean mask, remove the frequency components within the second preset frequency range in each frequency band respectively, to obtain a plurality of new abnormal vibration signals, so that the number of normal training samples is the same as the number of resonance fault training samples.

[0140] Specifically, in order to better train the deep learning model for detecting faults in marine diesel engines, the training samples of resonance faults with a small number of samples are augmented. The first preset frequency range in the abnormal vibration signal is divided into multiple frequency segments, and the first preset frequency range does not include the fault frequency. The frequency axis of each frequency segment is determined, and a boolean array with the same length as each frequency axis is generated respectively. Based on the second preset frequency range, a boolean mask for each boolean array is generated. Each time, according to the boolean mask, the frequency components within the frequency segment corresponding to the boolean mask are marked, and the frequency components within the frequency segment are set to zero.

[0141] Figure 5 Schematic diagrams of the original abnormal vibration signal and the abnormal vibration signal after rejection processing provided by the embodiments of this specification.

[0142] Taking 100 original normal vibration signals and 10 original abnormal vibration signals collected in the time domain as an example, if the first preset frequency range corresponding to each original abnormal vibration signal in the frequency domain is from 1 Hz to 100 Hz, in order to make the ratio of the number of normal training samples to the number of resonance fault training samples 1:1, the first preset frequency range corresponding to each original abnormal vibration signal can be divided into 10 equal-width frequency segments, and the width of each frequency segment is 10 Hz. According to the preset rules, the frequency components within the second preset frequency range are sequentially rejected in each frequency segment. Each rejection operation can generate a new abnormal vibration signal. For example, for the first time, the frequency components within the second preset frequency range in the first frequency segment are rejected from the original abnormal vibration signal, such as rejecting 1 Hz - 10 Hz, and for the second time, the frequency components within the second preset frequency range in the second frequency segment are rejected from the original abnormal vibration signal, such as rejecting 11 Hz - 20 Hz, and so on. The frequency components within the second preset frequency range in each frequency segment are sequentially rejected from the original abnormal vibration signal, and a total of 10 rejections are performed, and 10 new abnormal vibration signals can be obtained, as Figure 5 shown, 10 abnormal vibration signals can be obtained by rejecting the frequency components within the second preset frequency range from the original signal (Original Signal) respectively. According to the above method, 10 original abnormal vibration signals can be augmented to 100 new abnormal vibration signals, so that the ratio of normal training samples to resonance fault training samples is 1:1, and balance can be achieved.

[0143] In practical applications, other rejection methods can also be adopted. For example, different frequency components can be rejected in the same frequency band. For example, the frequency range of the first frequency band in the original abnormal vibration signal is 1 Hz - 10 Hz. For the first time, 1 Hz - 5 Hz in the first frequency band is rejected from the original abnormal vibration signal, and other frequency components are retained to generate a new sample. For the second time, 6 Hz - 10 Hz in the first frequency band is rejected from the original abnormal vibration signal, and other frequency components are retained to generate another new sample.

[0144] The second preset frequency range can be of fixed width or non-fixed width. Still taking the frequency range of the first frequency band as 1 Hz - 10 Hz as an example, for the first time, 1 Hz - 4 Hz in the first frequency band is rejected from the original abnormal vibration signal, and other frequency components are retained to generate a new sample. For the second time, 5 Hz - 10 Hz in the first frequency band is rejected from the original abnormal vibration signal, and other frequency components are retained to generate another new sample.

[0145] Similarly, for the first time, 1 Hz - 10 Hz in the first frequency band can be rejected from the original abnormal vibration signal, and other frequency components are retained to generate a new sample. For the second time, 11 Hz - 15 Hz in the second frequency band is rejected from the original abnormal vibration signal, and other frequency components are retained to generate another new sample.

[0146] Step 230: Fuse the time-domain features and frequency-domain features of each normal vibration signal and each new abnormal vibration signal respectively, and use the obtained feature vectors after fusion to train the support vector machine.

[0147] Feature extraction is performed on the normal vibration signal and the new abnormal vibration signal respectively. The time-domain features and frequency-domain features of the normal vibration signal are extracted, and the time-domain features and frequency-domain features of the new abnormal vibration signal are extracted. The time-domain features and frequency-domain features of each vibration signal are fused to obtain a feature vector corresponding to each vibration signal.

[0148] The obtained feature vectors are used to train the support vector machine to optimize the classification performance of the support vector machine, and the trained support vector machine is obtained.

[0149] Step 240: After feature extraction of the vibration signal to be detected, input it into the trained support vector machine to obtain the detection result, and it can be determined whether the ship diesel engine has a fault and the type of the fault.

[0150] Analyze the test results. If there are 100 normal training samples and 10 faulty training samples, and 23 time-frequency features and frequency-domain features are extracted from each training sample, form a feature vector by combining the time-frequency features and frequency-domain features. The ratio of the training set to the test set is set at 8:2 for the experiment. The number of training rounds is set at 100 rounds, and the penalty parameter C of the support vector machine is set at 0.1. To ensure the effectiveness of the experiment, the experiment is conducted 10 times, and observe the confusion matrix results output by the test set when the support vector machine model converges in each experiment.

[0151] Figure 6 This is a schematic diagram of the confusion matrix for fault prediction when the sample quantity is unbalanced provided by the embodiment of this specification.

[0152] The predicted results are shown through the confusion matrix, as Figure 6 shown. From the results, it can be obtained that in the 4th experiment, the model misclassified 2 faulty training samples as normal classes on the test set (20 normal training samples and 2 faulty training samples); in the 5th experiment, the model misclassified 2 faulty training samples as normal classes on the test set (19 normal training samples and 3 faulty training samples); in the 8th experiment, the model misclassified 2 faulty training samples as normal classes on the test set (19 normal training samples and 3 faulty training samples); in the 10th experiment, the model misclassified 2 faulty training samples as normal classes on the test set (17 normal training samples and 5 faulty training samples).

[0153] It shows that in the resonance fault diagnosis of marine diesel engines, when the samples are unbalanced, the support vector machine performs poorly in fault identification, the prediction accuracy of the model for minority-class samples is low, and it is difficult to effectively identify the resonance faults of marine diesel engines.

[0154] Figure 7 This is a schematic diagram of the confusion matrix for fault detection when the sample quantity is balanced provided by the embodiment of this specification.

[0155] The number of normal state samples remains unchanged at 100, and the number of faulty state samples is expanded from 10 to 100 to balance the detection results of the classifier for normal state samples. Taking the confusion matrix after the model converges as an example, the number of normal training samples and the number of resonance fault training samples in the test set are the same. As Figure 7 shown. From the experimental results, it can be obtained that the model can accurately predict the faulty training samples without misclassification, and the accuracy of the test set can reach 100%.

[0156] To verify the effect of the resonance fault diagnosis of marine diesel engines under the condition of adopting the processing method of marine diesel engine training samples in this application, use the accuracy Acc and the comprehensive index G-mean of unbalanced samples are used for comprehensive evaluation and comparison. The calculation formulas are as follows:

[0157]

[0158] Among them, TP, TN, FP, and FN respectively represent the number of positive-class samples with a positive detection result (normal state), the number of negative-class samples with a negative detection result (resonance fault state), the number of negative-class samples with a positive detection result, and the number of positive-class samples with a negative detection result.

[0159] Accuracy rate A cc It is used to measure the proportion of samples predicted correctly by the model in the total samples; the recall rate TPR is used to measure how many true positive-class samples the model can correctly identify; the true negative rate TNR is used to measure how many true negative-class samples the model can correctly identify; the value of G-mean is related to both majority-class and minority-class samples, and highlights the balance between the two through the geometric mean.

[0160] Figure 8 This is the evaluation result of the four evaluation indicators under the condition of unbalanced samples for the resonance fault diagnosis of marine diesel engines provided by the embodiments of this specification.

[0161] The accuracy rate A of the model test under the condition of unbalanced samples for the resonance fault diagnosis of marine diesel engines cc The three-line table display of the true negative rate TNR and the comprehensive index G-mean is the same as the output result of the confusion matrix. As Figure 8 shown, the evaluation indicators of the model in 10 experiments are unstable, which may lead to misjudgment and missed judgment by the staff during the actual fault diagnosis of marine diesel engines, and there is a certain risk.

[0162] Figure 9 This is the evaluation result of the four evaluation indicators under the condition of balanced samples for the resonance fault diagnosis of marine diesel engines provided by the embodiments of this specification.

[0163] The accuracy rate A of the model test under the condition of balanced samples for the resonance fault diagnosis of marine diesel engines cc The three-line table display of the true negative rate TNR and the comprehensive index G-mean is the same as the output result of the confusion matrix. As Figure 9 shown, the evaluation indicators of the model in 10 experiments can all be stabilized at 100%. Compared with the condition of unbalanced samples, it verifies the feasibility of the method for processing the training samples of marine diesel engines in this application.

[0164] Figure 10 This is the structural schematic diagram of a device for processing the training samples of marine diesel engines provided by the embodiments of this specification.

[0165] The processing device for the training samples of a marine diesel engine described in the embodiments of this specification may include:

[0166] An acquisition module 1002, configured to acquire a first vibration signal of a marine diesel engine in one acquisition cycle, where the first vibration signal includes abnormal signals;

[0167] A division module 1004, configured to divide a first preset frequency range corresponding to the first vibration signal into multiple frequency segments; the first preset frequency range does not include fault frequencies;

[0168] An elimination module 1006, configured to sequentially eliminate frequency components within a second preset frequency range in each of the frequency segments to obtain a second vibration signal after elimination processing corresponding to each of the frequency segments;

[0169] A building module 1008, configured to build a training sample set of the marine diesel engine based on all of the second vibration signals.

[0170] Optionally, in the embodiments of this specification, building the training sample set of the marine diesel engine based on all of the second vibration signals may specifically include:

[0171] Determining the time-domain features and frequency-domain features of each of the second vibration signals;

[0172] Fusing the time-domain features and the frequency-domain features of each of the second vibration signals respectively to obtain a feature vector corresponding to each of the second vibration signals;

[0173] Building the training sample set of the marine diesel engine based on all of the feature vectors.

[0174] Based on the same idea, the embodiments of this specification also provide a device corresponding to the above method.

[0175] Figure 11 It is a schematic structural diagram of a processing device for the training samples of a marine diesel engine provided by the embodiments of this specification. As Figure 11 shown, a processing device 1100 for the training samples of a marine diesel engine provided by the embodiments of this specification includes a memory 1130, a processor 1110, and a computer program 1120 stored in the memory. The processor 1110 executes the computer program 1120 to implement the processing method for the training samples of the marine diesel engine described in any of the above embodiments.

[0176] A processing device for the training samples of a marine diesel engine provided by the embodiments of this specification may include a memory, a processor, and a computer program stored in the memory. The processor executes the computer program to implement the processing method for the training samples of the marine diesel engine described in any of the above embodiments.

[0177] A computer-readable storage medium provided by an embodiment of the present specification, on which a computer program is stored, and when the computer program is executed by a processor, the processing method of the marine diesel engine training sample described in any one of the above embodiments can be implemented.

[0178] Each embodiment in this specification is described in a progressive manner. For the same or similar parts between the embodiments, reference can be made to each other. Each embodiment focuses on the differences from other embodiments. In particular, for Figure 11 the device shown, since it is basically similar to the method embodiment, the description is relatively simple, and for the relevant parts, reference can be made to the partial description of the method embodiment.

[0179] In the 1990s, it was obvious to distinguish whether an improvement to a technology was an improvement in hardware (e.g., improvement to circuit structures such as diodes, transistors, switches, etc.) or an improvement in software (improvement to method flows). However, with the development of technology, many improvements to method flows today can be regarded as direct improvements to hardware circuit structures. Almost all designers obtain the corresponding hardware circuit structure by programming the improved method flow into the hardware circuit. Therefore, it cannot be said that an improvement to a method flow cannot be implemented with a hardware entity module. For example, a Programmable Logic Device (PLD) (such as a Field Programmable Gate Array (FPGA)) is such an integrated circuit whose logical function is determined by the user's programming of the device. The designer can program by himself to "integrate" a digital system on a piece of PLD, without having to ask a chip manufacturer to design and produce a dedicated integrated circuit chip. Moreover, nowadays, instead of manually manufacturing integrated circuit chips, this programming is mostly implemented using "logic compiler" software, which is similar to the software compiler used in program development and writing. The original code before compilation also has to be written in a specific programming language, which is called a Hardware Description Language (HDL), and there is not only one kind of HDL, but many kinds, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc. The most commonly used ones currently are VHDL (Very-High-Speed Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should also be clear that as long as the method flow is slightly logically programmed with the above-mentioned several hardware description languages and programmed into the integrated circuit, it is easy to obtain the hardware circuit that implements the logical method flow.

[0180] The controller can be implemented in any suitable manner. For example, the controller can take the form of, for example, a microprocessor or a processor and a computer-readable medium storing computer-readable program code (such as software or firmware) executable by the (micro)processor, logic gates, switches, an application specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller. Examples of the controller include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320. The memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art also know that, in addition to implementing the controller in the form of pure computer-readable program code, it is entirely possible to make the controller implement the same function in the form of logic gates, switches, application specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, such a controller can be considered a hardware component, and the devices included therein for implementing various functions can also be regarded as the structures within the hardware component. Or even, the devices for implementing various functions can be regarded as either software modules for implementing the method or the structures within the hardware component.

[0181] The systems, devices, modules, or units illustrated in the above embodiments can be specifically implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer can be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or any combination of these devices.

[0182] For the convenience of description, when describing the above devices, they are described separately as various units according to their functions. Of course, when implementing the present application, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0183] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memories, CD-ROMs, optical memories, etc.) containing computer-usable program code.

[0184] The present invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It should be understood that each flow and / or block of the flowchart illustrations and / or block diagrams, and combinations of flows and / or blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions executed by the processor of the computer or other programmable data processing apparatus create means for implementing the functions specified in the flowchart flow or flows and / or block or blocks. Figure 1 in one flow or multiple flows and / or blocks Figure 1 or means for implementing the functions specified in one block or multiple blocks.

[0185] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture including instruction means that implement the functions specified in the flowchart flow or flows and / or block or blocks. Figure 1 in one flow or multiple flows and / or blocks Figure 1 or means for implementing the functions specified in one block or multiple blocks.

[0186] These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process, such that the instructions executed on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and / or block or blocks. Figure 1 in one flow or multiple flows and / or blocks Figure 1 or means for implementing the functions specified in one block or multiple blocks.

[0187] In a typical configuration, a computing device includes one or more processors (CPUs), an input / output interface, a network interface, and memory.

[0188] The memory may include non-permanent memory in the form of computer readable media, random access memory (RAM), and / or non-volatile memory, such as read only memory (ROM) or flash memory. The memory is an example of computer readable media.

[0189] A computer-readable medium includes both permanent and non-permanent, removable and non-removable media and can implement information storage by any method or technology. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic tape disk storage or other magnetic storage devices, or any other non-transitory medium that can be used to store information accessible by a computing device. As defined herein, a computer-readable medium does not include transitory computer-readable media such as modulated data signals and carrier waves.

[0190] It should also be noted that the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, such that a process, method, article or apparatus comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or apparatus. Without further limitation, an element defined by the phrase "comprising a..." does not exclude the presence of additional identical elements in the process, method, article or apparatus comprising said element.

[0191] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, system or computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0192] The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application can also be practiced in a distributed computing environment where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0193] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, various modifications and variations can be made to the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included within the scope of the claims of the present application.

Claims

1. A method for processing training samples of a marine diesel engine, characterized in that, Including: Obtain a first vibration signal of a marine diesel engine in one acquisition period, where the first vibration signal includes abnormal signals; Divide a first preset frequency range corresponding to the first vibration signal into multiple frequency segments; the first preset frequency range does not include fault frequencies; Successively eliminate frequency components within a second preset frequency range in each of the frequency segments to obtain a second vibration signal after elimination processing corresponding to each of the frequency segments; Based on all of the second vibration signals, establish a training sample set for the marine diesel engine.

2. The method according to claim 1, wherein The establishing the training sample set for the marine diesel engine based on all of the second vibration signals specifically includes: Determine the characteristic parameters of each of the second vibration signals; Based on the characteristic parameters, obtain a characteristic vector corresponding to each of the second vibration signals; Based on all of the characteristic vectors, establish a training sample set for the marine diesel engine.

3. The method according to claim 2, characterized in that, The characteristic parameters include time-domain characteristics and frequency-domain characteristics. The obtaining the characteristic vector corresponding to each of the second vibration signals based on the characteristic parameters, the method further includes: Fuse the time-domain characteristics and the frequency-domain characteristics of each of the second vibration signals respectively to obtain a characteristic vector corresponding to each of the second vibration signals.

4. The method according to claim 1, wherein Before the successively eliminating frequency components within a second preset frequency range in each of the frequency segments to obtain a second vibration signal after elimination processing corresponding to each of the frequency segments, the method further includes: Determine the frequency axis of each of the frequency segments; Generate a Boolean array with the same length as each of the frequency axes respectively; Based on the second preset frequency range, generate a Boolean mask for each of the Boolean arrays.

5. The method according to claim 4, characterized in that, The successively eliminating frequency components within a second preset frequency range in each of the frequency segments specifically includes: According to the Boolean mask, mark the frequency components within the frequency segment corresponding to the Boolean mask; Successively set the spectral amplitudes corresponding to the frequency components to zero.

6. The method according to claim 1, wherein The method further includes: Based on the training sample set, train a preset classification model to obtain a trained preset classification model.

7. A processing device for training samples of a marine diesel engine, characterized in that, Including: An acquisition module, configured to obtain a first vibration signal of a marine diesel engine in one acquisition period, where the first vibration signal includes abnormal signals; A division module, configured to divide a first preset frequency range corresponding to the first vibration signal into multiple frequency segments; the first preset frequency range does not include fault frequencies; An elimination module, configured to successively eliminate frequency components within a second preset frequency range in each of the frequency segments to obtain a second vibration signal after elimination processing corresponding to each of the frequency segments; A establishment module, configured to establish a training sample set for the marine diesel engine based on all of the second vibration signals.

8. The device according to claim 7, characterized in that, The establishing the training sample set for the marine diesel engine based on all of the second vibration signals specifically includes: Determine the time-domain characteristics and frequency-domain characteristics of each of the second vibration signals; Fuse the time-domain characteristics and the frequency-domain characteristics of each of the second vibration signals respectively to obtain a characteristic vector corresponding to each of the second vibration signals; Based on all of the characteristic vectors, establish a training sample set for the marine diesel engine.

9. A processing device for training samples of a marine diesel engine, comprising a memory, a processor, and a computer program stored on the memory, characterized in that: The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 6.

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

Images