Marine turbocharger state monitoring signal optimization and fault diagnosis method

Abstract

The invention discloses a marine turbocharger state monitoring signal optimization and fault diagnosis method, and the method comprises the steps: arranging a plurality of measurement points at a marine turbocharger, and collecting a vibration signal and a fault state corresponding to the vibration signal; carrying out validity identification on the vibration signal, wherein the validity identification comprises time domain detection and frequency domain detection; if the vibration signal is valid in both the time domain detection and the frequency domain detection, the vibration signal is valid data; variational mode decomposition is carried out on the IMF component, and part of the IMF component is selected to carry out signal reconstruction to obtain a reconstructed signal; calculating the improved information entropy of each reconstructed signal, and determining an optimal measurement point according to the improved information entropy; and effective time domain and frequency domain characteristic parameters are selected, the neural network model is trained by using the effective characteristic parameters of the vibration signals of the preferable measuring points and the corresponding fault states, and fault diagnosis of the marine turbocharger is completed. According to the method, screening of sensitive measuring points is completed based on the improved information entropy, and effective characteristic parameters are selected to complete fault diagnosis of the marine turbocharger.

Description

Technical Field

[0001] The invention relates to the technical field of engine marine turbocharger fault diagnosis, and in particular to a marine turbocharger state monitoring signal optimization and fault diagnosis method. Background Art

[0002] The monitoring, diagnosis and health status assessment of marine turbochargers can effectively assess and warn the actual operating status of the turbocharger, thereby preventing major accidents of the turbocharger. It is an important research direction for the design, operation and management of marine turbochargers. Its main processes include: the arrangement of sensors on the marine turbocharger body, vibration signal collection and storage, data validity judgment, vibration signal characteristic parameter calculation and analysis, diagnostic algorithm development, etc. At present, the research on fault diagnosis and health status assessment of marine turbochargers mainly has the following problems: (1) Before analyzing and processing the turbocharger vibration signal, the interference signal and noise signal were not removed, and the effective vibration signal was not identified. In the actual operation of the turbocharger, there are often many noises and interference signals. Noise mainly includes environmental white noise, etc. Interference signals mainly include the interference of the vibration signal of the marine engine on the turbocharger signal. If these signals are not removed and isolated, it will lead to misdiagnosis of turbocharger faults.

[0003] (2) Use a single sensor or a few types of signals for fault diagnosis. The vibration signal of the turbocharger contains fault information such as blade, rotor imbalance and bearing wear in the X, Y and Z directions. There are abnormal phenomena such as longitudinal vibration caused by axial movement, transverse vibration and flutter caused by the interaction between the blade and the fluid, and excessive vibration caused by abnormal bearing wear. At the same time, the energy of the measurement point signal is related to the distance from the turbocharger. Therefore, only arranging a single sensor and placing it in an improper position will result in insufficient fault information contained in the collected signal, resulting in poor fault diagnosis effect.

[0004] (3) When using data from multiple sensors for fault diagnosis, sensitive measurement points are not screened. Fault diagnosis requires signal fusion from multiple measurement points. If the measurement points are blindly arranged without screening sensitive measurement points, it will lead to too many measurement points and increased costs. At the same time, if data layer fusion from multiple measurement points is used, the data calculation rate will decrease due to the large amount of data, and the real-time performance will deteriorate. In addition, when collecting signals, the amount of effective information contained in sensors at different locations is different. Failure to screen sensitive measurement points may cause the final fusion features to contain too much interference information, which will greatly affect the accuracy of fault diagnosis.

[0005] (4) The characteristic laws of the turbocharger operating environment under variable speed conditions have not been analyzed. Due to the fluctuation and unevenness of engine combustion, and the influence of the propeller load on the ship propulsion system, which causes the diesel engine load to change, the speed of the turbocharger will change due to the change in exhaust gas energy, which in turn makes it difficult to distinguish the changing laws of the fault information contained in the vibration signal. For example, the fault signal at a lower speed and the fault-free vibration signal at a higher speed have great similarities in the time domain diagram, frequency domain diagram and their various characteristic parameters. If the characteristic parameters of the vibration signal of a single speed are directly applied to the variable speed condition, there is a problem of confusion between the fault characteristics and the normal characteristic signals.

[0006] In view of the above problems existing in the existing marine turbocharger fault diagnosis methods, it is necessary to study the optimization arrangement of marine turbocharger measurement points, signal optimization and anti-interference analysis, and diagnostic methods for variable operating conditions. Summary of the invention

[0007] The purpose of the present invention is to provide a method for optimizing a state monitoring signal and diagnosing a fault of a marine turbocharger, so as to at least solve the problems faced by the marine turbocharger fault diagnosis method, namely, difficulty in screening sensitive measuring points, lack of validity judgment of vibration signals, difficulty in selecting effective characteristic parameters and poor adaptability to variable speed conditions.

[0008] In a first aspect of the present invention, a method for optimizing state monitoring signals and diagnosing faults of a marine turbocharger is provided, the method comprising: Arrange multiple measuring points on the marine turbocharger and collect vibration signals and their corresponding fault conditions; Identify the validity of the collected vibration signals, including time domain detection and frequency domain detection; If the vibration signal is valid in both time domain detection and frequency domain detection, then the vibration signal is valid data; Perform variational mode decomposition on the vibration signal that is valid data to obtain IMF components of each order, and select some IMF components to reconstruct the signal to obtain the reconstructed signal; Calculate the improved information entropy of each reconstructed signal, and determine the optimal measurement point according to the improved information entropy; The time domain and frequency domain characteristic parameters of the vibration signal of the preferred measuring point are calculated, with the rotation speed as the horizontal coordinate and the characteristic parameter value as the vertical coordinate. A scatter plot is generated for each time domain and frequency domain characteristic parameter. Each scatter plot contains different fault states of the marine turbocharger, and different fault states use different marks. If the marks corresponding to different fault states in the scatter plot occupy an area respectively, the areas are clearly distinguished, and the marks corresponding to different fault states change smoothly with the speed, then the time domain and frequency domain feature parameters corresponding to the scatter plot are valid feature parameters; The neural network model is trained using the effective characteristic parameters of the vibration signal of the optimal measuring point and its corresponding fault state; The trained neural network model is subjected to single fault cross validation and multi-fault validation. The data used in the single fault cross validation are all data containing only one fault state, and the data used in the multi-fault validation are data containing multiple fault states. Including multiple fault states means that one data contains not only the fault state to be diagnosed, but also the fault state that does not need to be diagnosed, so as to verify whether the model can extract the fault state to be diagnosed from the multiple fault states; If the accuracy of single-fault cross-validation and multi-fault validation meets the requirements, the neural network model is qualified, and the marine turbocharger fault diagnosis is completed based on the qualified neural network model.

[0009] In some of the embodiments, a plurality of measuring points are arranged at different directions, different distances and different component positions of the marine turbocharger to collect vibration signals at different directions, different distances and different component positions of the marine turbocharger.

[0010] In some embodiments, there are 8 measuring points arranged as follows: The first to fifth measuring points are in the vertical Z direction, the seventh and eighth measuring points are in the X direction along the axial direction of the marine turbocharger rotor, and the sixth measuring point is in the Y direction perpendicular to the X and Z directions; wherein the X direction represents the longitudinal vibration caused by the axial movement of the turbocharger rotor, and the Y and Z directions represent the transverse vibration and flutter caused by the interaction between the blade and the fluid and the bearing wear; The first measuring point to the fourth measuring point is at a first distance from the turbocharger, the fifth measuring point to the eighth measuring point is at a second distance from the turbocharger, and the first distance is greater than the second distance; The first measuring point and the fourth to eighth measuring points are arranged at the compressor end of the turbocharger, and the second measuring point and the third measuring point are arranged at the turbine end of the turbocharger.

[0011] In some embodiments, the time domain detection includes: Determine whether the proportion of data points in the vibration signal whose voltage values ​​are less than or equal to the preset amplitude is greater than or equal to a first proportion; if so, the vibration signal is invalid; if not, perform least squares detrending term processing on the vibration signal, and calculate the effective value of the vibration signal after the detrending term, if the effective value exceeds the preset threshold, the vibration signal is invalid, and if the effective value does not exceed the preset threshold, the vibration signal is valid; Frequency domain detection includes: The spectrum of the vibration signal is obtained by Fourier transform. Combined with the turbocharger speed, the ratio of the frequency domain energy below the fundamental frequency of the turbocharger spectrum to the overall frequency domain energy is calculated. If the ratio is greater than or equal to the preset ratio, the vibration signal is invalid; otherwise, the vibration signal is valid.

[0012] In some embodiments, a least squares detrending term is performed on the vibration signal, including: Calculate b and a:

[0013]

[0014] In the formula, Indicates vibration signal, Indicates time, The number of data points representing the vibration signal; Calculate the vibration signal after detrending:

[0015] In the formula, Represents the vibration signal after detrending.

[0016] In some embodiments, some IMF components are selected to perform signal reconstruction to obtain a reconstructed signal, including: If the maximum measurement frequency of the vibration sensor is , each order IMF component contains The order is , then the 1st to the The IMF components of the order are removed; If the fundamental frequency of the turbocharger speed is , each order IMF component contains The order is The highest order of IMF components is selected as Step; The first Stage to The IMF components of the order are added together to obtain the reconstructed signal.

[0017] In some embodiments, calculating the improved information entropy of each reconstructed signal and determining the preferred measurement point according to the improved information entropy include: Calculate the information parameters of the frequency domain signal of the IMF component of the reconstructed signal:

[0018] In the formula, For the i The energy parameters of the IMF components, is the number of data points of the frequency domain signal, For the i The spectrum of the IMF component j The frequency value corresponding to the spectral line, for The corresponding amplitude; calculate The total energy of the IMF components :

[0019] In the formula, Represents the number of IMF components of the reconstructed signal; Calculating the probability of an event :

[0020]

[0021] Calculate the improved information entropy:

[0022]

[0023] In the formula, is the total number of speed intervals, k For the k Speed ​​range, Based on and Calculated k The improved information entropy of the speed range is Improved information entropy for the sum of all speed ranges; Multi-state improved information entropy average :

[0024] In the formula, Indicates the fault status number of the marine turbocharger, Indicates the number of marine turbochargers Fault state; Select The largest number of measuring points or The measuring point that exceeds the preset information entropy threshold is regarded as the optimal measuring point.

[0025] In some embodiments, the method further comprises: The vibration signal with valid data is saved as follows: Split the signal into multiple sub-signals according to the specified length; Add labels to each sub-signal, including speed, signal type, measurement point location and direction, and timestamp; signal types include vibration acceleration signal and axis trajectory signal; Add the fault status corresponding to the sub-signal to the label.

[0026] In some of the embodiments, the fault state of the marine turbocharger includes a normal state, a rotor slightly unbalanced state, and a rotor severely unbalanced state; The time domain and frequency domain characteristic parameters of the vibration signal include at least one of maximum value, mean value, absolute mean value, root amplitude, mean square value, effective value, variance, standard deviation, skewness, kurtosis, waveform factor, peak factor, impulse factor, margin index, kurtosis index, skewness index, amplitude mean value, center of gravity frequency, mean square frequency, frequency variance, root mean square frequency, amplitude variance and frequency standard deviation; A cubic polynomial curve is fitted to the mark corresponding to each fault state in the scatter plot. If the amplitude variation of the fitting curve in the entire speed range is less than the preset variation threshold, it is considered that the marks corresponding to different fault states in the scatter plot have a gentle change trend with the speed. The neural network model is a one-dimensional convolutional neural network; Before training the neural network model, the high-dimensional feature vector composed of the effective characteristic parameters of the vibration signal of the preferred measuring point is reduced in dimension by using the t-distributed random neighborhood embedding dimensionality reduction algorithm.

[0027] According to a second aspect of the present invention, a marine turbocharger is provided, which uses the marine turbocharger state monitoring signal optimization and fault diagnosis method described in any one of the first aspects to complete marine turbocharger fault diagnosis.

[0028] In general, the above technical solutions conceived by the present invention can achieve the following beneficial effects compared with the prior art: The present invention proposes a method for optimizing state monitoring signals and diagnosing faults of marine turbochargers. Through the design of a preliminary test scheme, identification and preservation of monitoring signal validity, optimization and arrangement of measuring points, research on multi-signal data fusion and diagnosis algorithms based on optimized measuring points, and a combination of a variable operating condition fault diagnosis algorithm library function package, the effective characteristic value selection, sensitive measuring point screening, fault diagnosis and health status assessment of marine turbochargers under variable speed conditions are achieved in a convenient, fast, high computational efficiency and high accuracy manner.

[0029] (1) The overall strategy of turbocharger fault detection test is proposed, including measurement point arrangement, data cleaning and data storage. The vibration signals of all directions of the turbocharger can be obtained relatively completely and representatively, and the effective signals can be saved as a directly usable data set.

[0030] (2) The calculation method of information entropy was improved and innovatively applied to the field of sensitive measuring point screening, realizing the screening of sensitive measuring points when arranging a large number of measuring points. At the same time, the proposed measuring point screening method can provide universal guidance for the arrangement of vibration sensor measuring points for marine turbochargers, which can reduce the number of measuring point arrangements and effectively reduce experimental costs.

[0031] (3) By analyzing the regularity of various characteristic parameters of vibration signals, a method is proposed to judge whether the characteristic parameters are effective through the speed-parameter scatter plot. This method can select effective characteristic parameters that are sensitive to faults and have a smaller trend of changing with speed.

[0032] (4) A dimensionality reduction algorithm was used to reduce the dimensionality of high-dimensional feature vectors, removing redundant information and improving computational efficiency. A convolutional neural network was then used to realize the recognition and classification of feature vectors with high accuracy.

[0033] Combined with the above effects, the marine turbocharger state monitoring signal optimization and fault diagnosis method proposed in the present invention analyzes and processes the marine turbocharger vibration signal, and realizes the fault diagnosis of the marine turbocharger. Compared with other existing methods, the marine turbocharger state monitoring signal optimization and fault diagnosis method proposed in the present invention has significant advantages in the sensitive measurement point screening effect and adaptability to variable speed working conditions. The method proposed in the present invention has high practical value for monitoring the operating status of marine turbochargers and ensuring the safe and stable operation of ship engine power systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 An overall flow chart of a method for optimizing state monitoring signals and diagnosing faults of a marine turbocharger provided in an embodiment of the present application; Figure 2 A sensor arrangement diagram provided in an embodiment of the present application; Figure 3 A flow chart for identifying the validity of a monitoring signal provided in an embodiment of the present application; Figure 4 A data storage flow chart provided for an embodiment of the present application; Figure 5 A flow chart of measuring point optimization provided in an embodiment of the present application; Figure 6 A schematic diagram of a data set provided in an embodiment of the present application; Figure 7 A flowchart of a multi-fault verification provided in an embodiment of the present application; Figure 8 A flow chart of speed change fault diagnosis provided in an embodiment of the present application; Fig. 9A flowchart of the development of variable operating condition fault diagnosis software provided in an embodiment of the present application. DETAILED DESCRIPTION

[0035] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. Based on the embodiments provided in this application, all other embodiments obtained by ordinary technicians in this field without making creative work belong to the scope of protection of the present invention.

[0036] Obviously, the drawings described below are only some examples or embodiments of the present application. For ordinary technicians in this field, the present application can also be applied to other similar scenarios based on these drawings without creative work. In addition, it can also be understood that although the efforts made in this development process may be complicated and lengthy, for ordinary technicians in this field related to the content disclosed in this application, some changes in design, manufacturing or production based on the technical content disclosed in this application are just conventional technical means, and should not be understood as insufficient content disclosed in this application.

[0037] Reference to "embodiments" in this application means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those of ordinary skill in the art that the embodiments described in this application may be combined with other embodiments without conflict.

[0038] Unless otherwise defined, the technical terms or scientific terms involved in this application should be understood by people with ordinary skills in the technical field to which this application belongs. The words "one", "a", "a", "the" and the like involved in this application do not indicate a quantitative limitation, and may represent the singular or plural. The terms "include", "comprise", "have" and any of their variations involved in this application are intended to cover non-exclusive inclusions; for example, a process, method, system, product or device that includes a series of steps or modules (units) is not limited to the listed steps or units, but may also include steps or units that are not listed, or may also include other steps or units inherent to these processes, methods, products or devices. The words "connect", "connected", "coupled" and the like involved in this application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The "multiple" involved in this application refers to two or more. "And / or" describes the association relationship of associated objects, indicating that there may be three relationships, for example, "A and / or B" can represent: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the objects before and after are in an "or" relationship. The terms "first", "second", "third", etc. involved in this application are only used to distinguish similar objects and do not represent a specific ordering of the objects.

[0039] The present application provides a method for arranging measuring points and optimizing signals for a marine turbocharger, as well as a method for judging the effectiveness of a signal and diagnosing a fault. The method is applicable to the design and development of products related to online fault diagnosis and health status assessment in terms of sensor arrangement and selection principles for monitoring marine turbochargers, signal acquisition analysis and effectiveness identification methods, intelligent diagnosis algorithms, and monitoring and diagnosis software.

[0040] Most existing marine turbocharger fault diagnosis methods face the problems of difficulty in screening sensitive measuring points, lack of effectiveness judgment of vibration signals, difficulty in selecting effective characteristic parameters, and poor adaptability to variable speed working conditions. The present invention proposes a marine turbocharger health status monitoring signal optimization and variable working condition fault diagnosis method, which selects effective characteristic parameters, screens sensitive measuring points, and develops intelligent diagnosis algorithms, solving the following problems: (1) In order to solve the problem of interference and noise in vibration signals, the online monitoring data of the supercharger was cleaned. According to the statistical characteristics of the vibration signals, the abnormal data points in the vibration signals were removed and replaced with reliable data points, thus improving the authenticity and reliability of the vibration signals. (2) To address the problem of insufficient measurement points and few signal types, vibration sensors are arranged in multiple locations and directions to collect vibration signals from multiple characteristic locations and representative directions of the turbocharger. The turbocharger fault diagnosis is performed in combination with the rotor axis trajectory, solving the problem of incomplete acquisition of fault characteristics due to single measurement points and data types. (3) To address the problem of blindly arranging and selecting measuring points, an improved information entropy calculation method was proposed, and a measuring point screening method was invented based on the improved information entropy. This method solves the adverse effects of information redundancy and interference signals on the diagnostic rate and accuracy when blindly selecting too many measuring point signals, reduces the number of sensors deployed, and saves costs. (4) In order to solve the problem that the existing algorithm is not adaptable to the speed, the characteristic parameter trend analysis of the turbocharger vibration signal at different speeds can be used to select effective characteristic parameters that are more sensitive to faults and have a more stable trend with speed changes, thus avoiding the problem that the characteristic parameter selection is not adaptable to the speed.

[0041] Finally, a method for single-fault-multiple-fault data model verification is designed, which provides a method to verify the fault diagnosis model from multiple perspectives such as accuracy, stability and generalization ability.

[0042] The marine turbocharger monitoring signal optimization and fault diagnosis method provided in this application is as follows: Figure 1 As shown, the process mainly includes five parts: design of the preliminary test plan, identification and preservation of the effectiveness of monitoring signals, optimization and layout of measuring points, research on multi-signal data fusion and diagnosis algorithms based on the optimized measuring points, and development of variable operating condition fault diagnosis algorithm library function packages. Through the combination of the above five parts, an efficient, convenient and high diagnostic accuracy marine turbocharger monitoring point optimization plan and diagnosis algorithm are formed, which realizes fault identification and health status assessment under the actual operation of the turbocharger with variable speed and anti-external interference. The specific technical solution is as follows: 1. Design of the preliminary test plan The arrangement of measuring points for the baseline test is as follows: Figure 2 As shown. The test is to arrange multiple measuring points around the turbocharger, and needs to include three directions: X, Y, and Z. The purpose is to obtain as many vibration signals as possible at different positions and directions of the turbocharger, and provide sufficient effective original signal references for subsequent signal processing, measuring point screening, etc. The sensor arrangement is as follows Figure 2As shown, measuring points 1 to 5 are in the vertical Z direction, measuring points 7 and 8 are in the X direction along the axial direction of the marine turbocharger rotor, and measuring point 6 is in the Y direction perpendicular to the X and Z directions. Among them, the X direction represents the longitudinal vibration caused by the axial movement of the turbocharger rotor, and the Y and Z directions represent the transverse vibration and flutter caused by the interaction between the blade and the fluid, bearing wear, etc.; measuring points 1-4 are far from the turbocharger, and measuring points 5-8 are close to the turbocharger. The use of a combination of far and near measuring point arrangement can verify the useful information contained in the turbocharger vibration signal at different propagation distances; measuring points 1, 4, 5, 6, 7, and 8 are arranged at the compressor end of the turbocharger, and measuring points 2 and 3 are arranged at the turbine end of the turbocharger. Measuring points are arranged at both the compressor end and the turbine end, which can verify the information contained in the turbocharger vibration signal at different components. Therefore, the developed method for arranging test measurement points, through the combination of different directions, different distances, and different component positions, enables the arranged 8 measurement points to obtain the vibration signals of the turbocharger more completely and representatively, laying the foundation for subsequent measurement point screening work.

[0043] 2. Identification and preservation of monitoring signal validity After the measurement points are arranged, the signals are collected, and the validity of the collected signals is identified and the data is saved. Figure 3 As shown in FIG, the effectiveness identification of the monitoring signal is divided into two parts: time domain detection and frequency domain detection.

[0044] Time domain detection is divided into three parts: (1) Check whether the time domain signal is very weak or even zero. Here, a weak signal is defined as the voltage value of 80% or more of the data points is less than or equal to 10mV. If this happens, it means that there is a problem with the corresponding signal acquisition channel and the vibration acceleration signal cannot be collected. The signal should be shielded or deleted, and the acquisition channel should be checked.

[0045] (2) To avoid signal drift, the signal is detrended. The signal is detrended according to the following formula: Perform least squares detrending process: ①Build the model:

[0046] ②Calculate parameters b and a:

[0047]

[0048] ③Calculate the signal after detrending :

[0049] (3) Calculate the effective value characteristic parameters of the signal after removing the trend term, and compare the calculated value with other adjacent measurement points in the same direction. If there are no adjacent measurement points in the same direction, compare it with historical data or empirical values. If the effective value exceeds the normal value range by more than 15%, the channel signal is judged as an abnormal signal and is shielded or deleted.

[0050] The frequency domain detection is as follows: Frequency domain detection obtains the spectrum of the time domain signal through Fourier transform, and calculates the ratio of the frequency domain energy below the fundamental frequency of the turbocharger spectrum to the overall frequency domain energy in combination with the turbocharger speed. If the ratio is greater than 10%, it is determined that the energy of the signal in the low-frequency area is unreasonable, and the signal of the channel is defined as an abnormal signal, which is shielded or deleted. The frequency domain energy below the fundamental frequency is calculated according to the following formula: ① Calculate turbocharger base frequency :

[0051] Where: is the speed of the turbocharger in rpm, The unit is .

[0052] ②Calculate the frequency domain energy below the fundamental frequency :

[0053] In the formula, is the frequency domain energy below the fundamental frequency, is the frequency of the signal, The maximum value is the turbocharger fundamental frequency , For the The amplitude corresponding to the spectral line.

[0054] If the signal passes the time domain detection and frequency domain detection, the signal is judged as a qualified signal to achieve data cleaning. After the data is cleaned, the valid data is saved. Data saving directly saves the data in the form of a usable data set to facilitate subsequent data calculation and fault diagnosis algorithm development.

[0055] like Figure 4 As shown, data storage is mainly divided into five parts: (1) Signal segmentation: The collected signal is divided into multiple sub-signals with each segment being 0.25 s in length.

[0056] (2) Add labels: Add measurement point and working condition information at the end of each data segment. The label must contain the following information: rotation speed, signal type (vibration acceleration signal, axis trajectory signal), measurement point position and direction, and timestamp.

[0057] (3) Distinguish the saving interval: Use the split and labeled signals for fault diagnosis. If the fault diagnosis result is normal, save the data every 60 minutes; if the fault diagnosis result is a fault, save the data every 1 minute; and add the fault diagnosis result to the data label. The data here is saved to a temporary database.

[0058] (4) Data sorting: The program cleans up the data every 24 hours and identifies the fault diagnosis label of the saved data. If the fault diagnosis label is normal, the redundant normal data is deleted according to the timestamp, and only the most recent normal data is retained; if the fault diagnosis label is faulty, all data is retained. The data here is saved to a long-term database, such as Figure 6 shown.

[0059] (5) After the data is sorted, clear the temporary database and prepare for the next save.

[0060] 3. Optimization and arrangement of measuring points like Figure 5 As shown in the figure, the measurement point optimization and arrangement module includes four parts: vibration signal decomposition, signal reconstruction, improved information entropy calculation, and sensitive measurement point selection: (1) Vibration signal decomposition The original vibration signal is subjected to variational mode decomposition (VMD) calculation, and each vibration signal is decomposed into multiple intrinsic mode functions (IMFs). The first layer is the component with the highest frequency in the vibration signal, and the last layer is the required lowest frequency component.

[0061] (2) Signal reconstruction selects the decomposed IMF components of the vibration signal. If the maximum measurement frequency of the vibration sensor is , each order IMF component contains The order is , then the 1st to the The purpose is to remove the high-frequency noise in the signal and the signal outside the sensor measurement range. If the fundamental frequency of the turbocharger speed is , each order IMF component contains The order is The highest order of IMF components is selected as The purpose is to remove the lower frequency interfering vibration signals, such as the vibration signals of the internal combustion engine, while retaining the vibration signals of the turbocharger at lower speeds as much as possible. After selecting the appropriate IMF order, the selected IMF components are added and summed to obtain the reconstructed signal.

[0062] The traditional method for calculating the energy entropy of vibration signals only uses the amplitude of the time domain or frequency domain signals of each order component, which are all the ordinate values ​​of the signal, but does not use the time or frequency value of its abscissa. This calculation method ignores the amount of information in one dimension of the signal. Using this method, it is impossible to fully calculate the total amount of information of the signal. Therefore, it is necessary to improve the traditional energy entropy calculation method and add the frequency value of the abscissa of the frequency domain signal to the information entropy calculation process, that is, to improve the information entropy.

[0063] (3) Improved information entropy calculation The improved information entropy of the reconstructed signal at each measuring point is calculated according to the following five steps to evaluate the amount of information contained in the signal collected at each measuring point.

[0064] ① Calculate the information parameters of the frequency domain signal of the IMF component:

[0065] Where: For the i The energy parameters of the IMF components, is the number of data points of the frequency domain signal, For the i The spectrum of the IMF component j The frequency value corresponding to the spectral line, for The corresponding amplitude.

[0066] ②Calculation m The total energy of the IMF components:

[0067] ③Calculate the probability of an event:

[0068]

[0069] ④ Calculate the improved information entropy:

[0070]

[0071] Where: k is the rotational speed. For example, there are 11 rotational speeds from 35000 rpm to 60000 rpm (the step interval of the rotational speed when collecting data is 2500 rpm), so the value of k is 1 to 11. Based on and Calculated k The improved information entropy of the speed range is Improve the information entropy for the sum of all speed ranges.

[0072] ⑤ Take the average value of multi-state information entropy: In order to avoid the randomness and instability of the information entropy value of a single fault state, it is necessary to take the average value of the information entropy of all fault states. For example, if there are three states: normal state, slight rotor imbalance, and severe rotor imbalance, then the information entropy of these three states is averaged:

[0073] (4) Selection of Sensitive Measurement Points The information entropy of each measurement point obtained in the improved information entropy calculation is sorted, and the four measurement points with the largest information entropy are selected as sensitive measurement points.

[0074] The sensitive measuring points selected by this method can be used as the measuring point layout locations for collecting vibration signals of marine turbochargers.

[0075] 4. Research on multi-signal data fusion and diagnosis algorithm based on optimal measurement points like Figure 8 As shown in the figure, the multi-signal data fusion and diagnosis algorithm research module based on the optimal measurement point includes five parts: characteristic parameter calculation, characteristic parameter trend analysis, effective characteristic parameter selection, characteristic vector dimension reduction and fault diagnosis: (1) Characteristic parameter calculation The common time domain and frequency domain characteristic parameters of the marine turbocharger vibration signal are calculated using the formula. The calculation method is shown in Table 1.

[0076] Table 1 Characteristic parameter calculation method table

[0077] (2) Characteristic parameter trend analysis: According to the data saved by the data storage module, the time domain and frequency domain characteristic parameters of each sub-signal are calculated using the characteristic parameter calculation method described in Table 1, and the changing trend of the characteristic parameters is analyzed using a scatter plot. The method is as follows: the horizontal axis is the rotation speed, the vertical axis is the characteristic parameter value, and each characteristic parameter generates a scatter plot. Each scatter plot should contain the characteristic values ​​of signals with different fault levels (for example: normal state, slightly unbalanced rotor state, and severely unbalanced rotor state). The scatter points of each fault level use different colors. The principles for judging the effectiveness of characteristic parameters based on scatter plots are as follows: ① In a scatter plot, each color of scatter points occupies an area, and scatter points of different colors are clearly separated; ② Under the premise of meeting condition ①, fit the scattered points of each color with a cubic polynomial curve. If the amplitude variation of the fitting curve of the scattered points of each color in the entire speed range is less than 15%, it is considered that the characteristic parameter changes smoothly with the speed, and the characteristic parameter represented by the scatter plot can be selected as a valid characteristic parameter. The amplitude variation of the fitting curve is calculated according to the following formula:

[0078] In the formula, is the amplitude variation of the fitting curve, is the maximum value of the fitting curve, is the minimum value of the fitted curve.

[0079] Through this method, the effective characteristic parameters selected are the centroid frequency, mean square frequency, and root mean square frequency.

[0080] (3) Selection of effective characteristic parameters The centroid frequency, mean square frequency, and root mean square frequency selected in the characteristic parameter trend analysis are summarized as effective characteristic parameters.

[0081] (4) Feature vector dimensionality reduction: The high-dimensional feature vector composed of the effective feature parameters of the measuring points selected in the sensitive measuring point selection is subjected to dimensionality reduction processing by using a t-distributed random neighborhood embedding dimensionality reduction algorithm to reduce the high-dimensional feature vector to three dimensions; (5) Data classification: A one-dimensional convolutional neural network is used to classify and diagnose the feature parameters after dimensionality reduction. The diagnosis is divided into two parts: single fault cross-validation and multi-fault validation.

[0082] The data used in the single-fault cross-validation are all data containing only one type of fault, such as normal state data, rotor slight imbalance fault data, and rotor severe imbalance fault data.

[0083] After the single-fault cross-validation is completed, if the average accuracy is ≥85%, it is judged as qualified.

[0084] After the single-fault cross-validation is qualified, in order to further verify the robustness and versatility of the fault diagnosis algorithm, multi-fault data can be further used for fault diagnosis verification.

[0085] The data used for multi-fault verification contains data on multiple faults, such as normal state data, rotor minor imbalance fault data, and rotor severe imbalance + bearing wear dual fault data. When performing multi-fault verification, the neural network model should not be retrained, but the neural network model trained with single fault data should be used directly. The purpose of using multi-fault data for fault diagnosis is to verify whether the algorithm used is versatile, that is, whether the fault to be diagnosed can be extracted from the multi-fault data, for example: whether the rotor severe imbalance + bearing wear dual fault data can be used to diagnose a rotor severe imbalance fault, rather than diagnosing it as a normal signal or a rotor minor imbalance fault.

[0086] After the multi-fault verification is completed, if the average accuracy rate is ≥85%, it is judged as qualified.

[0087] The process of multi-fault verification is as follows Figure 7shown.

[0088] If the single fault cross validation is qualified, it proves that the algorithm has a high accuracy rate; if both the single fault cross validation and the multi-fault validation are qualified, it proves that the algorithm has not only a high accuracy rate, but also high robustness and versatility. This verification method can provide a reference for the effect evaluation of the fault diagnosis model.

[0089] 5. Development of fault diagnosis software for variable working conditions like Fig. 9 As shown in the figure, the fault diagnosis software for variable working conditions uses the MATLAB software platform to develop the fault diagnosis algorithm, and uses the virtual instrument LabVIEW software to develop the monitoring and diagnosis software. The developed monitoring and diagnosis software includes signal acquisition parameter setting module, signal acquisition module, data display module, fault diagnosis module, etc.

[0090] When using LabVIEW for fault diagnosis, it is necessary to use the algorithm script import function of the LabVIEW software to integrate the fault diagnosis algorithm developed using MATLAB into LabVIEW, and realize the integrated process of signal acquisition and fault diagnosis based on virtual instruments through the mixed programming method of M and G languages.

[0091] After using LabVIEW for software development, the project may contain multiple sub-files (VIs). At this time, the related VIs need to be packaged into the form of LabVIEW library function packages (lvlib). Through this method, the related VIs can be managed in an orderly manner. In addition, Ivlib allows setting access rights. The access rights of VIs can be freely set to specify which VIs are public (Public) and which are private (Private). This helps to protect sub-VIs that do not want to be called directly and improve the security of the code.

[0092] In addition, the monitoring and diagnostic software project developed by LabVIEW needs to be integrated and packaged to eventually generate an executable file (.exe file) so that the monitoring and diagnostic software can run on a computer that does not have LabVIEW and MATLAB software installed.

[0093] In summary, the present application provides a method for optimizing the status monitoring signal and diagnosing faults of a marine turbocharger. Through the design of a preliminary test scheme, identification and preservation of the effectiveness of monitoring signals, optimization and arrangement of measuring points, research on multi-signal data fusion and diagnostic algorithms based on optimized measuring points, and a combination of a variable operating condition fault diagnosis algorithm library function package, the present application realizes the selection of effective characteristic values, screening of sensitive measuring points, fault diagnosis and health status assessment of marine turbochargers under variable speed conditions in a convenient, fast, high computing efficiency and high accuracy manner.

[0094] (1) The overall strategy of turbocharger fault detection test is proposed, including measurement point arrangement, data cleaning and data storage. The vibration signals of all directions of the turbocharger can be obtained relatively completely and representatively, and the effective signals can be saved as a directly usable data set.

[0095] (2) The calculation method of information entropy was improved and innovatively applied to the field of sensitive measuring point screening, realizing the screening of sensitive measuring points when arranging a large number of measuring points. At the same time, the proposed measuring point screening method can provide universal guidance for the arrangement of vibration sensor measuring points for marine turbochargers, which can reduce the number of measuring point arrangements and effectively reduce experimental costs.

[0096] (3) By analyzing the regularity of various characteristic parameters of vibration signals, a method is proposed to judge whether the characteristic parameters are effective through the speed-parameter scatter plot. This method can select effective characteristic parameters that are sensitive to faults and have a smaller trend of changing with speed.

[0097] (4) A dimensionality reduction algorithm was used to reduce the dimensionality of high-dimensional feature vectors, removing redundant information and improving computational efficiency. A convolutional neural network was then used to realize the recognition and classification of feature vectors with high accuracy.

[0098] (5) Through the hybrid programming method of MATLAB and LabVIEW, fault diagnosis algorithms and data monitoring and fault diagnosis software were developed, and the relevant subVIs were integrated into the lvlib library file package to enhance the manageability of project files. The final data monitoring and fault diagnosis software was compiled into an executable file, realizing an integrated process of data acquisition and fault diagnosis based on virtual instruments.

[0099] Combined with the above effects, the marine turbocharger fault diagnosis method proposed in the present invention analyzes and processes the marine turbocharger vibration signal, thereby realizing the fault diagnosis of the marine turbocharger. Compared with other existing methods, the marine turbocharger fault diagnosis method proposed in the present invention has significant advantages in the screening effect of sensitive measuring points and adaptability to variable speed working conditions. The method proposed in the present invention has high practical value for monitoring the operating status of marine turbochargers and ensuring the safe and stable operation of ship engine power systems.

[0100] It should be pointed out that the technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, all possible combinations of the technical features in the above-mentioned embodiments are not described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification. In addition, according to the needs of implementation, the various steps / components described in this application can be split into more steps / components, and two or more steps / components or partial operations of steps / components can be combined into new steps / components to achieve the purpose of the present invention.

[0101] It is easy for a person skilled in the art to understand that the above-mentioned embodiments only express several implementation methods of the present application, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the invention patent. It should be pointed out that for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A method for optimizing state monitoring signals and diagnosing faults of a marine turbocharger, characterized in that: The method includes: Arrange multiple measuring points on the marine turbocharger and collect vibration signals and their corresponding fault conditions; Identify the validity of the collected vibration signals, including time domain detection and frequency domain detection; If the vibration signal is valid in both time domain detection and frequency domain detection, then the vibration signal is valid data; Perform variational mode decomposition on the vibration signal that is valid data to obtain IMF components of each order, and select some IMF components to reconstruct the signal to obtain the reconstructed signal; Calculate the improved information entropy of each reconstructed signal, and determine the optimal measurement point according to the improved information entropy; The time domain and frequency domain characteristic parameters of the vibration signal of the preferred measuring point are calculated, with the rotation speed as the horizontal coordinate and the characteristic parameter value as the vertical coordinate. A scatter plot is generated for each time domain and frequency domain characteristic parameter. Each scatter plot contains different fault states of the marine turbocharger, and different fault states use different marks. If the marks corresponding to different fault states in the scatter plot occupy an area respectively, the areas are clearly distinguished, and the marks corresponding to different fault states change smoothly with the speed, then the time domain and frequency domain feature parameters corresponding to the scatter plot are valid feature parameters; The neural network model is trained using the effective characteristic parameters of the vibration signal of the optimal measuring point and its corresponding fault state; The trained neural network model is subjected to single fault cross validation and multi-fault validation. The data used in the single fault cross validation are all data containing only one fault state, and the data used in the multi-fault validation are data containing multiple fault states. Including multiple fault states means that one data contains not only the fault state to be diagnosed, but also the fault state that does not need to be diagnosed, so as to verify whether the model can extract the fault state to be diagnosed from the multiple fault states; If the accuracy of single-fault cross-validation and multi-fault validation meets the requirements, the neural network model is qualified, and the marine turbocharger fault diagnosis is completed based on the qualified neural network model.

2. The method for optimizing the state monitoring signal and diagnosing the fault of a marine turbocharger according to claim 1, characterized in that: Multiple measuring points are arranged at different directions, different distances and different component positions of the marine turbocharger to collect vibration signals at different directions, different distances and different component positions of the marine turbocharger.

3. The method for optimizing state monitoring signals and diagnosing faults of a marine turbocharger according to claim 2, characterized in that: There are 8 measuring points, arranged as follows: The first to fifth measuring points are in the vertical Z direction, the seventh and eighth measuring points are in the X direction along the axial direction of the marine turbocharger rotor, and the sixth measuring point is in the Y direction perpendicular to the X and Z directions; wherein the X direction represents the longitudinal vibration caused by the axial movement of the turbocharger rotor, and the Y and Z directions represent the transverse vibration and flutter caused by the interaction between the blade and the fluid and the bearing wear; The first measuring point to the fourth measuring point is at a first distance from the turbocharger, the fifth measuring point to the eighth measuring point is at a second distance from the turbocharger, and the first distance is greater than the second distance; The first measuring point and the fourth to eighth measuring points are arranged at the compressor end of the turbocharger, and the second measuring point and the third measuring point are arranged at the turbine end of the turbocharger.

4. The method for optimizing state monitoring signals and diagnosing faults of a marine turbocharger according to claim 1, characterized in that: Time domain detection includes: Determine whether the proportion of data points in the vibration signal whose voltage values ​​are less than or equal to the preset amplitude is greater than or equal to a first proportion; if so, the vibration signal is invalid; if not, perform least squares detrending term processing on the vibration signal, and calculate the effective value of the vibration signal after the detrending term, if the effective value exceeds the preset threshold, the vibration signal is invalid, and if the effective value does not exceed the preset threshold, the vibration signal is valid; Frequency domain detection includes: The spectrum of the vibration signal is obtained by Fourier transform. Combined with the turbocharger speed, the ratio of the frequency domain energy below the fundamental frequency of the turbocharger spectrum to the overall frequency domain energy is calculated. If the ratio is greater than or equal to the preset ratio, the vibration signal is invalid; otherwise, the vibration signal is valid.

5. The method for optimizing state monitoring signals and diagnosing faults of a marine turbocharger according to claim 4, characterized in that: The vibration signal is processed by least squares detrending term, including: Calculate b and a: In the formula, Indicates vibration signal, Indicates time, The number of data points representing the vibration signal; Calculate the vibration signal after detrending: In the formula, Represents the vibration signal after detrending.

6. The method for optimizing state monitoring signals and diagnosing faults of a marine turbocharger according to claim 1, characterized in that: Select some IMF components to reconstruct the signal and obtain the reconstructed signal, including: If the maximum measurement frequency of the vibration sensor is , each order IMF component contains The order is , then the 1st to the The IMF components of the order are removed; If the fundamental frequency of the turbocharger speed is , each order IMF component contains The order is The highest order of IMF components is selected as Step; The first Stage to The IMF components of the order are added together to obtain the reconstructed signal.

7. The method for optimizing state monitoring signals and diagnosing faults of a marine turbocharger according to claim 1, characterized in that: Calculate the improved information entropy of each reconstructed signal and determine the optimal measurement point according to the improved information entropy, including: Calculate the information parameters of the frequency domain signal of the IMF component of the reconstructed signal: In the formula, For the i The energy parameters of the IMF components, is the number of data points of the frequency domain signal, For the i The spectrum of the IMF component j The frequency value corresponding to the spectral line, for The corresponding amplitude; calculate The total energy of the IMF components : In the formula, Represents the number of IMF components of the reconstructed signal; Calculating the probability of an event : Calculate the improved information entropy: In the formula, is the total number of speed intervals, k For the k Speed ​​range, Based on and Calculated k The improved information entropy of the speed range is Improved information entropy for the sum of all speed ranges; Multi-state improved information entropy average : In the formula, Indicates the fault status number of the marine turbocharger, Indicates the number of marine turbochargers Fault state; Select The largest number of measuring points or The measuring point that exceeds the preset information entropy threshold is regarded as the optimal measuring point.

8. The method for optimizing state monitoring signals and diagnosing faults of a marine turbocharger according to claim 1, characterized in that: The method further includes: The vibration signal with valid data is saved as follows: Split the signal into multiple sub-signals according to the specified length; Add labels to each sub-signal, including speed, signal type, measurement point position and direction, and timestamp; signal types include vibration acceleration signal and axis trajectory signal; Add the fault status corresponding to the sub-signal to the label.

9. The method for optimizing state monitoring signals and diagnosing faults of a marine turbocharger according to claim 1, characterized in that: The fault conditions of marine turbochargers include normal condition, slight rotor imbalance condition and severe rotor imbalance condition; The time domain and frequency domain characteristic parameters of the vibration signal include at least one of maximum value, mean value, absolute mean value, root amplitude, mean square value, effective value, variance, standard deviation, skewness, kurtosis, waveform factor, peak factor, impulse factor, margin index, kurtosis index, skewness index, amplitude mean value, center of gravity frequency, mean square frequency, frequency variance, root mean square frequency, amplitude variance and frequency standard deviation; A cubic polynomial curve is fitted to the mark corresponding to each fault state in the scatter plot. If the amplitude variation of the fitting curve in the entire speed range is less than the preset variation threshold, it is considered that the marks corresponding to different fault states in the scatter plot have a gentle change trend with the speed. The neural network model is a one-dimensional convolutional neural network; Before training the neural network model, the high-dimensional feature vector composed of the effective characteristic parameters of the vibration signal of the preferred measuring point is reduced in dimension by using the t-distributed random neighborhood embedding dimensionality reduction algorithm.

10. A marine turbocharger, characterized in that: The marine turbocharger adopts the marine turbocharger state monitoring signal optimization and fault diagnosis method described in any one of claims 1 to 9 to complete the marine turbocharger fault diagnosis.

Images