CN117647758A - AUV propeller state monitoring system and method based on pre-detection - Google Patents
AUV propeller state monitoring system and method based on pre-detection Download PDFInfo
- Publication number
- CN117647758A CN117647758A CN202410123163.2A CN202410123163A CN117647758A CN 117647758 A CN117647758 A CN 117647758A CN 202410123163 A CN202410123163 A CN 202410123163A CN 117647758 A CN117647758 A CN 117647758A
- Authority
- CN
- China
- Prior art keywords
- module
- function
- data
- wavelet
- propeller
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 67
- 238000012544 monitoring process Methods 0.000 title claims abstract description 47
- 238000001514 detection method Methods 0.000 title claims abstract description 33
- 238000012545 processing Methods 0.000 claims abstract description 30
- 238000004891 communication Methods 0.000 claims abstract description 17
- 239000004973 liquid crystal related substance Substances 0.000 claims abstract description 15
- 238000013500 data storage Methods 0.000 claims abstract description 11
- 230000006870 function Effects 0.000 claims description 58
- 230000009466 transformation Effects 0.000 claims description 19
- 230000008569 process Effects 0.000 claims description 13
- 238000004458 analytical method Methods 0.000 claims description 12
- 230000007246 mechanism Effects 0.000 claims description 11
- 239000011159 matrix material Substances 0.000 claims description 10
- 230000015654 memory Effects 0.000 claims description 9
- 230000005540 biological transmission Effects 0.000 claims description 8
- 238000013519 translation Methods 0.000 claims description 7
- 238000004364 calculation method Methods 0.000 claims description 6
- 238000000354 decomposition reaction Methods 0.000 claims description 6
- 238000007781 pre-processing Methods 0.000 claims description 6
- 230000000007 visual effect Effects 0.000 claims description 5
- 230000001413 cellular effect Effects 0.000 claims description 3
- 238000001914 filtration Methods 0.000 claims description 3
- 238000009499 grossing Methods 0.000 claims description 3
- 239000003550 marker Substances 0.000 claims description 3
- 238000007726 management method Methods 0.000 description 22
- 108091006146 Channels Proteins 0.000 description 18
- 238000012423 maintenance Methods 0.000 description 11
- 238000010586 diagram Methods 0.000 description 6
- 230000008859 change Effects 0.000 description 5
- 238000003745 diagnosis Methods 0.000 description 5
- QVFWZNCVPCJQOP-UHFFFAOYSA-N chloralodol Chemical compound CC(O)(C)CC(C)OC(O)C(Cl)(Cl)Cl QVFWZNCVPCJQOP-UHFFFAOYSA-N 0.000 description 4
- 238000010606 normalization Methods 0.000 description 4
- 230000004913 activation Effects 0.000 description 3
- 230000006978 adaptation Effects 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 3
- 230000006399 behavior Effects 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 3
- 238000011835 investigation Methods 0.000 description 3
- 238000011176 pooling Methods 0.000 description 3
- 238000012216 screening Methods 0.000 description 3
- 241000669244 Unaspis euonymi Species 0.000 description 2
- 238000006243 chemical reaction Methods 0.000 description 2
- 238000013523 data management Methods 0.000 description 2
- 238000007418 data mining Methods 0.000 description 2
- 230000001419 dependent effect Effects 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 230000003993 interaction Effects 0.000 description 2
- 230000007787 long-term memory Effects 0.000 description 2
- 230000007774 longterm Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000008447 perception Effects 0.000 description 2
- 230000004044 response Effects 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 1
- 230000003213 activating effect Effects 0.000 description 1
- 238000007792 addition Methods 0.000 description 1
- 239000000853 adhesive Substances 0.000 description 1
- 230000001070 adhesive effect Effects 0.000 description 1
- 238000012550 audit Methods 0.000 description 1
- 230000002457 bidirectional effect Effects 0.000 description 1
- 239000000872 buffer Substances 0.000 description 1
- 238000012790 confirmation Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000012217 deletion Methods 0.000 description 1
- 230000037430 deletion Effects 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 239000006185 dispersion Substances 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005265 energy consumption Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 229910052500 inorganic mineral Inorganic materials 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 239000011707 mineral Substances 0.000 description 1
- 230000002035 prolonged effect Effects 0.000 description 1
- 238000011084 recovery Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012876 topography Methods 0.000 description 1
Landscapes
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The invention provides an AUV propeller state monitoring system and method based on pre-detection, which belongs to the technical field of autonomous underwater robot propellers, wherein the system comprises: the system comprises a signal acquisition unit, a signal processing unit, a main control and interface module, a wireless data communication module, a data storage unit, an upper computer, a liquid crystal display module and an alarm module; the signal acquisition unit acquires current, voltage and rotating speed signals of the propeller, the current, voltage and rotating speed signals are sent to the signal processing unit through the main control and interface module, the liquid crystal display module is used for displaying the signals in real time, the main control and interface module is connected with the data storage unit and used for storing historical data acquired from the signal acquisition unit and the processing unit, the historical data and the upper computer are mutually transmitted through the wireless data communication module, and the upper computer is used for carrying out data query or parameter setting. The technical scheme of the invention solves the problem of low fault identification precision caused by insufficient state monitoring of the AUV propeller in the prior art.
Description
Technical Field
The invention relates to the technical field of Autonomous Underwater Vehicle (AUV) propeller, in particular to an AUV propeller state monitoring system and method based on pre-detection.
Background
Unmanned Underwater Vehicles (UUV) mainly refer to intelligent systems for replacing divers or manned small submarines to perform deep sea exploration, mine investigation and other high-risk underwater operations, and are mainly divided into remote-control underwater Robots (ROVs) and autonomous underwater robots (AUVs) according to cabled and cableless conditions. AUV is widely applied to marine science research, submarine resource exploration and military investigation, and can be used for monitoring a marine ecosystem and tracking marine organisms, water quality and topography at the bottom of the ocean; the method can also execute efficient exploration tasks and find potential oil and gas resources or mineral points; the system can be further used for underwater information collection, target searching and reconnaissance tasks, and becomes a key tool for military operations and offshore safety.
The propeller is a key power system of the AUV, and directly influences the maneuverability and operability of the AUV. The propeller system with good performance can ensure that the AUV can freely move in water for a long time, execute various complex work tasks and improve the working efficiency. However, since AUV operation is in a complex and severe deep sea environment, extreme environmental conditions such as high pressure, low temperature, high salinity and the like require that the propeller have high-strength reliability. Once the propeller breaks down, the motor performance of the AUV is likely to be lost, a series of chain reactions are generated, sensor data is lost, other equipment is damaged, and the AUV is even out of control and cannot float. Therefore, establishing an AUV propeller state online monitoring system has important significance in maintaining the safety, reliability and high efficiency of an AUV system.
The on-line monitoring system can track the running state of the propeller in real time, discover and diagnose faults and problems in time, and is beneficial to taking corresponding decision-making and maintenance measures; and the energy consumption can be reduced by optimizing the operation parameters, and the service lives of the propeller and the whole AUV system can be prolonged. In addition, the large amount of data collected by the online monitoring system provides a basis for a data-driven maintenance strategy. Through machine learning and data mining techniques, the system can learn and adapt to performance of the propeller under different working conditions, and a more accurate maintenance plan is formulated.
Thus, there is a need for an AUV propeller state monitoring system and method based on pre-detection.
Disclosure of Invention
The invention mainly aims to provide an AUV propeller state monitoring system and method based on pre-detection, which are used for solving the problem of low fault identification precision caused by insufficient AUV propeller state monitoring in the prior art.
To achieve the above object, the present invention provides an AUV propeller state monitoring system based on pre-detection, including: the system comprises a signal acquisition unit, a signal processing unit, a main control and interface module, a wireless data communication module, a data storage unit, an upper computer, a liquid crystal display module and an alarm module; the signal acquisition unit acquires current, voltage and rotating speed signals of the propeller, and sends the signals to the signal processing unit through the main control and interface module, wherein the signal processing unit comprises a preprocessing method and all fault monitoring methods; the main control and interface module is connected with the liquid crystal display module and the alarm module, the liquid crystal display module is used for displaying current, voltage and rotating speed signals in real time, and when the output result of the internal signal processing unit is a fault, the alarm module is controlled to send out an audible and visual signal alarm; the main control and interface module is connected with the data storage unit and used for storing the history data acquired from the signal acquisition unit and the processing unit, and the data transmission is carried out between the main control and interface module and the upper computer through the wireless data communication module, and the upper computer carries out data query or parameter setting.
The invention also provides an AUV propeller state monitoring method based on pre-detection, which specifically comprises the following steps:
s1, collecting current, voltage and rotating speed signals of a propeller.
S2, preprocessing the acquired signals, including: filtering, smoothing and aligning data to obtain a denoising signal.
S3, pre-detection is carried out, and the extracted time domain characteristic parameters comprise dimensional parameters and dimensionless parameters.
And S4, setting a threshold according to a 3 sigma rule, continuously adjusting the threshold, and accurately detecting when the time domain characteristic parameter in one period exceeds the threshold.
S5, for the signals exceeding the threshold value in the step S4, the signals are decomposed into different frequency bands and sub-frequency bands by utilizing wavelet packet transformation.
S6, constructing an LSTM long-short-term memory model LSTM-MANN with a multi-scale convolution and an attention mechanism by utilizing fault data of the historical propeller, and inputting time-frequency characteristics after wavelet packet transformation processing into the LSTM-MANN model;
and S7, inputting the data processed in the step S5 into an LSTM-MANN model, generating a real-time status report and fault early warning information of the propeller according to a fault recognition result output by the LSTM-MANN model, and transmitting the real-time status report and the fault early warning information to an upper computer through wireless communication.
Further, the step S3 specifically includes:
calculating the dimensional parameters includes: arithmetic mean, peak-to-peak, standard deviation, root mean square, kurtosis, and skewness.
Calculating dimensionless parameters includes: waveform factor, peak factor, kurtosis factor, pulse factor, margin factor and skewness factor.
Further, the calculation formula of the threshold in step S4 is:
（1）；
（2）；
wherein,representing the probability that the probability is to be determined,represents the average value of the values,represents the standard deviation of the standard,represents a threshold value and,representing a data point of the data,representing the total number of data points.
Further, the step S5 specifically includes the following steps:
s5.1 for a square integrable functionI.e.If Fourier transformThe method meets the following conditions:
（3）。
wherein the square integrable functionIs a wavelet mother function and is also called a wavelet function as shown in the formula (3)Is derived from the allowable conditions:
square integrable functionThe following must be satisfied:the method comprises the steps of carrying out a first treatment on the surface of the And needs to satisfy:to ensure thatAt the position ofNearby continuity, whereIn order to be able to take time,is a positive real number.
S5.2, defining wavelet mother functionThe time window width of (2) is，Is the frequency domain window width of。
S5.3, wavelet mother functionTo extend and retract and translate, provided withIs of the size factor ofTranslation factor ofMake the function after expansion and translation asThe following steps are:
（4）；
in the method, in the process of the invention,to be dependent on continuously variable parametersAndis used for the wavelet basis functions of (1).
S5.4 for signalsThe following steps are:
（5）；
（6）；
wherein,a space representing all square integrable functions;for the normalization constant(s),is a wavelet transform coefficient.
According to the fourier transform product theorem, equation (6) is written as an equivalent frequency domain expression form:
（7）；
in the method, in the process of the invention,is thatIs used for the fourier transform of (a),is thatHarmonic components of different frequencies are used,is a plurality of the components of the liquid crystal display,is the frequency.
Further, step S5 further includes the steps of:
s5.5, setting the orthogonal scale function and the wavelet mother function in wavelet packet transformation as respectivelyAndthe low-pass filter in the filter bank isThe high-pass filter isAccording to the theory of multi-resolution analysis,、、、the two-scale equation satisfied between is:
（8）。
in the formula, the coefficient sequenceAndthe method meets the following conditions:
（9）。
s5.6, introducing a marker:；
definition:
： （10）；
to be formed by orthogonal scale functionsGenerating a general iteration formula of the wavelet packet basis function;as discrete time index, whenIn the time-course of which the first and second contact surfaces,i.e. scale functionThe method comprises the steps of carrying out a first treatment on the surface of the When (when)In the time-course of which the first and second contact surfaces,i.e. wavelet function。
S5.7, setRepresenting signalsIn the first placeLayer on layer 1The wavelet packet is called wavelet packet coefficient, and the decomposition algorithm of wavelet packet transformation is as follows:
（11）；
in the method, in the process of the invention,；；；low-pass and high-pass filters, respectively, in wavelet decomposition, respectively, with scale functionsAnd wavelet functionRelated to the following.
The reconstruction algorithm of the wavelet packet transformation is as follows:
（12）。
further, step S6 includes the steps of:
s6.1, constructing an LSTM-MACNN structure, wherein the LSTM structure comprises an LSTM module, three multi-level multi-scale attention convolution modules, namely Section-1, section-2 and Section-3, and a classification module.
S6.2, LSTM Module comprising an input GateThe corresponding weight matrix isThe method comprises the steps of carrying out a first treatment on the surface of the Forgetting doorThe corresponding weight matrix isThe method comprises the steps of carrying out a first treatment on the surface of the Output doorThe corresponding weight matrix is；
（13）。
Wherein,in order to be in a temporary cellular state,for the current state of the memory cell,is the hidden state of the next cell unit.
The invention has the following beneficial effects:
the invention combines a time domain feature pre-detection method and a wavelet packet transformation enhanced LSTM-MACNN model in the field of AUV propeller state on-line monitoring. The threshold analysis method is primarily used for screening the time domain characteristic parameters so as to realize the pre-detection of faults, quickly identify key statistical characteristics in the operation data of the propeller, and provide primary insight for deep analysis. After that, the introduction of wavelet packet transformation provides the LSTM-MACNN model with richer and finer multi-scale input features, greatly enhances the capability of the model to capture long-short-term dependence characteristics in time sequence data, and enables the system to show higher efficiency and accuracy in the aspect of accurate fault detection of the AUV propeller. In addition, the software system designed by the invention provides a comprehensive platform for operators through the modularized design and the man-machine interaction interface, and greatly enhances the practicability and reliability of the AUV propeller state on-line monitoring system.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are needed in the description of the embodiments or the prior art will be briefly described, and it is obvious that the drawings in the description below are some embodiments of the present invention, and other drawings can be obtained according to the drawings without inventive effort for a person skilled in the art. In the drawings:
fig. 1 shows a block diagram of an AUV propeller condition monitoring system based on pre-detection of the present invention.
Fig. 2 shows a schematic circuit diagram of the system of the present invention.
Fig. 3 shows a flow chart of an AUV propeller state monitoring method based on pre-detection of the present invention.
FIG. 4 is a flowchart showing the LSTM-MACNN model construction method of step S6 of the present invention.
Fig. 5 shows a software system structure diagram provided by the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made apparent and fully in view of the accompanying drawings, in which some, but not all embodiments of the invention are shown. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
An AUV propeller state monitoring system based on pre-detection as shown in fig. 1, comprising: the system comprises a signal acquisition unit, a signal processing unit, a main control and interface module, a wireless data communication module, a data storage unit, an upper computer, a liquid crystal display module and an alarm module; the signal acquisition unit acquires current, voltage and rotating speed signals of the propeller, and sends the signals to the signal processing unit through the main control and interface module, wherein the signal processing unit comprises a preprocessing method and all fault monitoring methods; the main control and interface module is connected with the liquid crystal display module and the alarm module, the liquid crystal display module is used for displaying current, voltage and rotating speed signals in real time, and when the output result of the internal signal processing unit is a fault, the alarm module is controlled to send out an audible and visual signal alarm; the main control and interface module is connected with the data storage unit and used for storing the history data acquired from the signal acquisition unit and the processing unit, and the data transmission is carried out between the main control and interface module and the upper computer through the wireless data communication module, and the upper computer carries out data query or parameter setting. Specifically, as shown in fig. 2, the main control and interface module adopts an STM32 core controller module, specifically adopts an STM32F407ZGT6 chip, is responsible for logic control and data processing of the whole system, and outputs a PWM signal to the PWM driving module through a GPIO port thereof to control the rotation speed change of the brushless motor; the signal acquisition unit includes: the data of the rotating speed, current and voltage acquisition modules are received through the I2C interface of the STM32F407ZGT6 chip; the system is communicated with a liquid crystal display module LCD, and the system state information which is acquired and processed by the signal acquisition unit is displayed; and when the output result of the signal processing unit is a fault, the control alarm module sends out an audible and visual signal alarm. The data storage unit, i.e. the SD card, is connected to the STM32F407ZGT6 chip.
Specifically, the AUV propeller state monitoring system also includes a power module that provides a stable power supply to the STM32 core controller module, brushless motors, sensors, and other electronic components.
The brushless motor receives a drive signal from the PWM drive module of the STM32F407ZGT6 chip, and adjusts the rotation speed according to the change of the signal.
The PWM driving module receives PWM signals from the STM32 core controller module and adjusts the working state of the brushless motor.
The motor state real-time monitoring module comprises a rotating speed acquisition module, a current acquisition module, a voltage acquisition module, a liquid crystal display module and an alarm module, wherein: the rotation speed acquisition module adopts an AS5600 magnetic encoder chip, is a 12-bit high-resolution magnetic rotation absolute position sensor, can output a change range of 0 to 360 degrees under default conditions, and provides an I2C interface. The chip being non-contacting with the magnet of the measuring medium when measuring angle, i.e. the magnet being in contact withThere is 1-3 mm's clearance between the chip, and this magnet passes through the back output shaft of adhesive with the motor to link to each other, through the rotation of motor company area magnet, causes the inside perception structure of chip to produce corresponding change, and then carries out the angle measurement. The AS5600 chip is internally provided with a configuration register, an output register and a status register, and the registers can be subjected to read-write operation through I2C, wherein an Angle register is more important and is used for storing real-time Angle information; and the Status register is used for storing the magnetic field intensity information of the chip working environment. In addition, a system click timer in a HAL library of the STM32 main control chip is utilized to acquire system time and count value, and the time difference between two sampling is calculated. In order to deal with the situation that the jump is 0 after the angle exceeds 360 degrees, the positive and negative and the range of the angle in positive and negative rotation are determined by using a quadrant dividing form, and the rotation number is determined by combining the change of the relative angle, so that the rotation speed at the current moment is calculatedThe calculation formula is as follows:
。
wherein,represents the number of rotations and,representing the relative angle of the two parts of the lens,representing the time difference. The current acquisition module and the voltage acquisition module adopt INA219 chips which are provided with I2C interfaces, are bidirectional current controller chips, are embedded with delta-sigma ADC, can convert input analog voltage and divided voltage signals into digital signals, and can be stored in a voltage register and a divided voltage register, and can convert the divided voltage signals into current signals through an internal circuit and can be stored in the current register. The module is connected with the analog input endEnd and its useThe terminals are connected with the two ends of a sampling resistor in the circuit, and input voltage is input through a selection channel and an analog-digital converterAnd partial pressureThe current is calculated by combining the divided voltage and a calibration value, and the calculation formula is as follows:
；
；
；
in the method, in the process of the invention,is a high-precision resistor voltage;is the bus voltage;、、the values of the current register, the shunt voltage register and the calibration register are respectively.
The LCD module adopts MCU resistance screen to display the collected information of rotation speed, current, voltage, etc. and the result of pre-detection and accurate detection.
The alarm module adopts a mode of combining a buzzer and an LED lamp, is activated when fault information is detected after the signal processing of the core controller module, and performs audible and visual alarm.
The wireless data communication module adopts a LoRa module, is connected with the singlechip through a wireless module interface, packages data into LoRa information, and sends the LoRa information to the upper computer through wireless transmission. The upper computer end is provided with corresponding receiving equipment or gateway for receiving the data transmitted by the LoRa module.
As shown in fig. 3, the present invention further provides a detection method for monitoring the state of an AUV propeller based on pre-detection, which specifically includes the following steps:
s1, collecting current, voltage and rotating speed signals of a propeller.
S2, preprocessing the acquired signals, including: filtering, smoothing and aligning data to obtain a denoising signal.
S3, pre-detection is carried out, and the extracted time domain characteristic parameters comprise dimensional parameters and dimensionless parameters.
And S4, setting a threshold according to a 3 sigma rule, continuously adjusting the threshold, and accurately detecting when the time domain characteristic parameter in one period exceeds the threshold.
S5, for the signals exceeding the threshold value in the step S4, the signals are decomposed into different frequency bands and sub-frequency bands by utilizing wavelet packet transformation, so that richer input features are provided for the neural network.
S6, constructing an LSTM long-short-term memory model LSTM-MANN with a multi-scale convolution and attention mechanism by utilizing the fault data of the history thruster, and inputting the time-frequency characteristics after wavelet packet transformation processing into the LSTM-MANN model.
And S7, inputting the data processed in the step S5 into an LSTM-MANN model, generating a real-time status report and fault early warning information of the propeller according to a fault recognition result output by the LSTM-MANN model, and transmitting the real-time status report and the fault early warning information to an upper computer through wireless communication.
Specifically, step S3 specifically includes:
calculating the dimensional parameters includes: arithmetic mean, peak-to-peak, standard deviation, root mean square, kurtosis, and skewness.
(1) Arithmetic mean value: mean level of data is represented for understanding general trends in current, voltage or rotational speed.
。
(2) Peak value: refers to the absolute value of the highest point for determining the maximum intensity of the current, voltage or rotational speed.
。
(3) Peak-to-peak value: the difference between the highest and lowest points is used to describe the total amplitude of the signal.
。
(4) Standard deviation of: the degree of dispersion of the data distribution is measured for evaluating the stability of the current, voltage or rotational speed.
。
(5) Root mean square: also known as effective value, is commonly used to represent the strength of a signal, and is one of the important indicators for monitoring the running state of a propeller.
。
(6) Kurtosis of: describing how sharp the data is distributed, the kurtosis means that the data has more extreme values.
。
(7) Skewness of inclination: symmetry of the data distribution is measured to determine whether current, voltage or speed readings tend to be higher or lower values.
。
Calculating dimensionless parameters includes: waveform factor, peak factor, kurtosis factor, pulse factor, margin factor and skewness factor.
(1) Waveform factor: for describing the degree of deviation of the signal waveform with respect to the sine wave.
。
(2) Peak factor: for detecting whether an impact is present in the signal.
。
(3) Kurtosis factor: reactive current loadingThe degree of flattening of the waveform.
。
(4) Pulse factor: for identifying extreme peaks in the signal, facilitating analysis of the ability to cope with extreme loads.
。
(5) Margin factor: a margin of the indicator circuit or system relative to the maximum rated condition under normal operating conditions, a high margin factor means that the system has a greater fault tolerance.
。
(6) Bias factor: the degree of asymmetry of the response signal relative to the center.
。
Specifically, according to the 3 sigma rule, as in equation 1, about 99.7% of the data values should fall within ±3 sigma of the mean, values outside of this range are generally considered outliers or outliers. The calculation formula of the threshold in step S4 is:
（1）；
（2）；
wherein,representing the probability that the probability is to be determined,represents the average value of the values,represents the standard deviation of the standard,represents a threshold value and,representing a data point of the data,representing the total number of data points.
Specifically, further accurate detection is enabled for those types of signals exceeding the threshold in step S4. Firstly, wavelet packet transformation is carried out on the preprocessed signals, and through the finer time-frequency analysis method, not only low-frequency components reflecting the overall trend of the signals, but also high-frequency components of each frequency section for describing main detail information of the signals are obtained, so that key features in the signals are more comprehensively captured. The step S5 specifically comprises the following steps:
s5.1 for a square integrable functionI.e.If Fourier transformThe method meets the following conditions:
（3）。
wherein the square integrable functionIs a wavelet mother function and is also called a wavelet function as shown in the formula (3)Is derived from the allowable conditions:
square integrable functionThe following must be satisfied:the method comprises the steps of carrying out a first treatment on the surface of the And needs to satisfy:to ensure thatAt the position ofNearby continuity, whereIn order to be able to take time,is a positive real number.
S5.2, defining wavelet mother functionThe time window width of (2) is，Is the frequency domain window width of。
S5.3, wavelet mother functionTo extend and retract and translate, provided withIs of the size factor ofTranslation factor ofMake the function after expansion and translation asThe following steps are:
（4）；
in the method, in the process of the invention,to be dependent on continuously variable parametersAndis used for the wavelet basis functions of (1). Inverse of scale factor and frequencyThere is a certain relationship: small scale corresponds to high frequencies and large scale corresponds to low frequencies. If the scale of the wavelet mother function is compared with a time window, the readily available small scale corresponds to the short-time signal one by one, and the large scale corresponds to the long-time signal one by one, so that the method accords with the natural rule of time-frequency distribution of the signal, namely, the high-frequency signal has short duration and the low-frequency signal has long duration.
S5.4 for signalsThe following steps are:
（5）；
（6）；
wherein,a space representing all square integrable functions;for the normalization constant(s),is a wavelet transform coefficient.
According to the fourier transform product theorem, equation (6) is written as an equivalent frequency domain expression form:
（7）；
in the method, in the process of the invention,is thatIs used for the fourier transform of (a),is thatHarmonic components of different frequencies are used,is a plurality of the components of the liquid crystal display,is the frequency.
Specifically, step S5 further includes the steps of:
s5.5, setting the orthogonal scale function and the wavelet mother function in wavelet packet transformation as respectivelyAndthe low-pass filter in the filter bank isThe high-pass filter isAccording to the theory of multi-resolution analysis,、、、the two-scale equation satisfied between is:
（8）。
in the formula, the coefficient sequenceAndthe method meets the following conditions:
（9）。
s5.6, introducing a marker:；
definition:
： （10）；
to be formed by orthogonal scale functionsGenerating a general iteration formula of the wavelet packet basis function;as discrete time index, whenIn the time-course of which the first and second contact surfaces,i.e. scale functionThe method comprises the steps of carrying out a first treatment on the surface of the When (when)In the time-course of which the first and second contact surfaces,i.e. wavelet function。
S5.7, setRepresenting signalsIn the first placeLayer (dimension)) Upper firstThe wavelet packet is called wavelet packet coefficient, and the decomposition algorithm of wavelet packet transformation is as follows:
（11）；
in the method, in the process of the invention,；；；low-pass and high-pass filters, respectively, in wavelet decomposition, respectively, with scale functionsAnd wavelet functionRelated to the following.
The reconstruction algorithm of the wavelet packet transformation is as follows:
（12）。
specifically, an LSTM long-short-term memory (LSTM-MACNN) model with a multi-scale convolution and attention mechanism is constructed, time-frequency characteristics after wavelet packet transformation processing are input into the model, firstly, the LSTM module is used for extracting the dependence characteristics of each time step of monitoring data, then the multi-scale attention convolution module is stacked behind the LSTM module, residual connection is added, the multi-scale convolution is enabled to extract complex local characteristics of different scales of the monitoring data and strong dependence characteristics after LSTM processing, and finally the importance degree of each channel characteristic diagram is learned through the channel attention mechanism. The simulated test data is then divided into two classes, a portion of which serves as a training set to train the model's ability to identify and classify the normal and abnormal states of the propeller, and the remainder serves as a test set to verify the model training effect.
Step S6 includes the steps of:
s6.1, constructing an LSTM-MACNN structure, wherein the LSTM structure comprises an LSTM module, three multi-level multi-scale attention convolution modules, namely Section-1, section-2 and Section-3, and a classification module. Wherein the LSTM module includes: two LSTM networks, a residual connection structure and a batch normalization layer BN, two LSTM network layers for capturing and analyzing time point correlation features in the monitored data, including long-term and short-term time dependency features. In order to enhance the feature extraction capability of the model and ensure the stability of the gradient flow, a residual error connection structure is introduced after two LSTM layers, which is beneficial to the effective fusion of the original data and the processed data. And then accessing a batch normalization layer BN for normalizing the input features and maintaining the distribution consistency of the network so as to accelerate the convergence of the model.
Next, LSTM-MACNN contains three multi-level Section modules, section-1, section-2, and Section-3. The first two layers of each Section module are multi-scale attention convolution (MAC) layers, and each MAC layer is composed of a multi-scale convolution module and an attention module and is used for extracting local features of different scales and highlighting key information. But the ends are slightly different, wherein the Section-1 and the Section-2 are configured with the largest pooling layer, and the aim of screening out the most distinguishing characteristics is achieved; and the tail end of the Section-3 is configured with a global average pooling layer, and the generated feature map is directly used for classification and assisted with a Dropout mechanism to reduce the risk of overfitting.
And finally, configuring a classification layer at the tail part of the whole model, and outputting probability distribution of each type of trend through a Softmax function to realize accurate classification of the monitoring data. In general, the model can improve the processing efficiency and classification accuracy of on-line monitoring data while ensuring the depth and complexity, and meets the high-precision requirement of AUV propeller state monitoring on data analysis and processing.
The long-term memory neural network (LSTM) is an improvement of the cyclic neural network (RNN), and long-term memory is realized by screening important signals, so that the problem of gradient disappearance of the standard RNN when the standard RNN is used for treating long-term dependence is solved.
S6.2，LSThe TM module includes an input gateThe corresponding weight matrix isThe method comprises the steps of carrying out a first treatment on the surface of the Forgetting doorThe corresponding weight matrix isThe method comprises the steps of carrying out a first treatment on the surface of the Output doorThe corresponding weight matrix is；
（13）。
In order to be in a temporary cellular state,for the current state of the memory cell,is the hidden state of the next cell unit. These gates are all set to output a real vector of 0 to 1. Using current inputsHidden state generated at last momentAnd the current memory state of the cellFor determining whether to accept input, before forgettingThe stored memory and the output of the next generated state. When the gate output is 0, it indicates that any information cannot pass; when the gate output is 1, it means that any information can pass without limit.
The MAC layer in fig. 4 contains a multi-scale convolution module and an attention module. Wherein the multi-scale convolution module comprises 4 layers. The first layer is a convolution layer, three parts of one-dimensional convolution products with different sizes are adopted to carry out convolution operation on the input feature images, time sequence features with different scales are extracted, and corresponding feature images are generated. The multi-scale convolution kernel can generate different-sized perception fields, the sizes of the three parts of convolution kernels are sequentially increased, the internal dependence characteristics of the short, medium and long subsequences are extracted, and the short, medium and long dependence characteristics of the propeller online monitoring data are correspondingly obtained. The second layer is a cascade layer, the three feature images with different sizes are spliced in the channel dimension, and the number of channels of the spliced feature images isThen. The number of input channels isThe number of channels of each convolution kernel of the first layer isThe second cascade layer is. And outputting the spliced characteristic diagram to a BN layer for batch standardization processing. The last layer is an activation layer, and ReLU is used as an activation function to improve the nonlinear expression capacity of the model and reduce the risk of overfitting.
The mechanism of attention is a mechanism for studying human vision and mainly comprises two aspects: firstly, focusing on important information; and secondly, different weights are given to the heavy point information and the noise information. The mechanism can screen key characteristic information from multiple parameters of the propeller, filter redundant information and realize multiple characteristicsAnd optimizing the weight parameters, so as to improve the diagnosis precision. In the present invention, the input of the attention module is the output of the multi-scale convolution moduleThe output is。
The attention module contains 3 layers. The first layer is global average pooling layer, and features are mappedConversion to a vector matrixThe calculation formula is as follows:
；
in the method, in the process of the invention,is the firstThe feature map of each channel is used for extruding global time sequence information into a channel descriptor by calculating the average value of data in the feature map of each channel;is the firstDescriptors of the individual channels represent statistical information.
The last two layers are all connection layers for obtaining channel weights. In the first full connection layer, the dimension-reducing factor is utilizedCompressing the number of channels toTo achieve channel interaction, then atReducing the number of channels to the second full connection layerObtaining channel weightsThe method comprises the following steps:
；
in the method, in the process of the invention,activating a function for a ReLU;the weight parameters of the 2 full connection layers are respectively;for sigmoid activation functions, channel weights can be mapped between (0, 1), the more the weights are close to 1, the more feature information the trend identifies, the more important the channel feature, and vice versa. Finally, the characteristic diagram is inputAnd weight ofAnd performing point multiplication operation to obtain Output of the MAC layer.
Fig. 5 is a functional design plan of a software system provided by the present invention, which includes a system management module, an equipment management module, a communication management module, a diagnostic system module, an area management module, a data management module, and a maintenance management module.
The system management module is configured with user management, rights management, and behavior log functions. The user management function is responsible for account creation, modification, deletion and password management; the authority management function controls the access of users of different levels to system functions and data by setting different access authorities; the behavior log function is used for recording the operation behavior of a user and the running state of the system, and is convenient for system audit and fault investigation.
The device management module is configured with concentrator management, terminal management functions. Wherein the concentrator manages the summary, forwarding and possibly preliminary analysis for the data, and also monitors the integrity and performance of the data flow, and buffers and backups the data if necessary; the terminal management function is used for directly managing and maintaining each monitoring terminal, and also comprises parameter configuration, state monitoring, fault diagnosis and maintenance updating of terminal equipment, so that the correct operation and data accuracy of each monitoring point are ensured.
The communication management module is configured with a data transmission management function, a communication protocol adaptation function, and a secure encryption mechanism. The data transmission management function is responsible for coordinating and managing data exchange between the inside of the system and external equipment, and ensuring timeliness and accuracy of data transmission; the communication protocol adaptation function supports diversified communication standards and protocols, thereby realizing compatibility with various types of devices and networks; the security encryption mechanism protects the security of data in the transmission process through an advanced encryption algorithm and a security protocol, and prevents unauthorized access and data tampering.
The diagnostic system module is configured with threshold settings, historical status queries, real-time data display, and online assessment diagnostic functions. Wherein the threshold setting function allows a user to set a threshold for the monitoring parameter according to experience or statistical data for judging the state of the equipment; the historical state query function allows a user to query historical operating data and state records of the propeller and related devices; the real-time data display function displays the current data in real time in the form of charts, curves and the like, so that a user is helped to quickly capture the state of the equipment; the on-line evaluation and diagnosis function combines the real-time data and the history experience to perform on-line evaluation and fault diagnosis on the equipment state.
The regional management module allows a user to conduct regional management on the monitoring system according to different geographic positions or task regions; the grouping of the equipment and the data in the region is supported, so that the regional performance analysis and maintenance management are convenient; and customization of the regional monitoring strategy is provided, and the flexibility of the specific regional monitoring requirement is met.
The data management module is configured with a real-time data processing function, a historical data storage function, and a data backup and restore function. The real-time data processing function processes and stores the real-time data collected by the sensor for instant analysis and display; the historical data storage function is used for storing the running data of the propeller for a long time and supporting subsequent historical trend analysis and data mining; the data backup and recovery function ensures the integrity of critical data to prevent system failure or data loss.
The maintenance management module is configured with an early warning information confirmation and maintenance plan setting function. The early warning information confirming function allows a user to confirm and manage early warning information sent by the diagnosis system module, so that the accuracy of faults and the timely response of the system are ensured; the maintenance plan setting function supports the user to make and adjust the maintenance plan of the equipment, predictive maintenance and execution condition tracking thereof.
In summary, the invention provides a method and a system for on-line monitoring of the state of an autonomous underwater robot propeller based on time domain feature pre-detection and deep learning accurate detection, which relate to threshold analysis of time domain statistical parameters and LSTM-MACNN model fault detection based on wavelet packet transformation enhancement, and effectively improve the accuracy and efficiency of AUV propeller fault detection.
It should be understood that the above description is not intended to limit the invention to the particular embodiments disclosed, but to limit the invention to the particular embodiments disclosed, and that the invention is not limited to the particular embodiments disclosed, but is intended to cover modifications, adaptations, additions and alternatives falling within the spirit and scope of the invention.
Claims (7)
1. An AUV propeller state monitoring system based on pre-detection, comprising: the system comprises a signal acquisition unit, a signal processing unit, a main control and interface module, a wireless data communication module, a data storage unit, an upper computer, a liquid crystal display module and an alarm module;
the signal acquisition unit acquires current, voltage and rotating speed signals of the propeller, and sends the signals to the signal processing unit through the main control and interface module, and the signal processing unit comprises preprocessing and all fault monitoring methods; the main control and interface module is connected with the liquid crystal display module and the alarm module, the liquid crystal display module is used for displaying current, voltage and rotating speed signals in real time, and when the output result of the internal signal processing unit is a fault, the alarm module is controlled to send out an audible and visual signal alarm; the main control and interface module is connected with the data storage unit and used for storing the history data acquired from the signal acquisition unit and the processing unit, and the data transmission is carried out between the main control and interface module and the upper computer through the wireless data communication module, and the upper computer carries out data query or parameter setting.
2. An AUV propeller state monitoring method based on pre-detection, using the system of claim 1, comprising the steps of:
s1, collecting current, voltage and rotating speed signals of a propeller;
s2, preprocessing the acquired signals, including: filtering, smoothing and aligning data to obtain a denoising signal;
s3, pre-detection is carried out, and time domain characteristic parameters including dimensional parameters and dimensionless parameters are extracted;
s4, setting a threshold according to a 3 sigma rule, continuously adjusting the threshold, and accurately detecting when the time domain characteristic parameter in one period exceeds the threshold;
s5, for the signals exceeding the threshold value in the step S4, decomposing the signals into different frequency bands and sub-frequency bands by utilizing wavelet packet transformation;
s6, constructing an LSTM long-short-term memory model LSTM-MACNN with a multi-scale convolution and an attention mechanism by utilizing the fault data of the historical propeller;
and S7, inputting the data processed in the step S5 into an LSTM-MANN model, generating a real-time status report and fault early warning information of the propeller according to a fault recognition result output by the LSTM-MANN model, and transmitting the real-time status report and the fault early warning information to an upper computer through wireless communication.
3. The method for monitoring the state of an AUV propeller based on pre-detection according to claim 2, wherein step S3 specifically comprises:
calculating the dimensional parameters includes: arithmetic mean, peak-to-peak, standard deviation, root mean square, kurtosis, and skewness;
calculating dimensionless parameters includes: waveform factor, peak factor, kurtosis factor, pulse factor, margin factor and skewness factor.
4. The method for monitoring the state of an AUV propeller based on pre-detection according to claim 3, wherein the calculation formula of the threshold in step S4 is:
（1）；
（2）；
wherein,representing probability->Represents mean value>Represents standard deviation (S)>Represents threshold value->Representing data points, +.>Representing the total number of data points.
5. The method for monitoring the state of an AUV propeller based on pre-detection according to claim 4, wherein the step S5 specifically comprises the steps of:
s5.1 for a square integrable functionI.e. +.>If Fourier transform->The method meets the following conditions:
（3）；
wherein the square integrable functionIs a wavelet mother function and is also called as a wavelet function +.>Is derived from the allowable conditions:
square integrable functionThe following must be satisfied: />The method comprises the steps of carrying out a first treatment on the surface of the And needs to satisfy:to ensure->At->Continuity of the vicinity, wherein->For time of arrivalBetween (I) and (II)>Is a positive real number;
s5.2, defining wavelet mother functionIs +.>，/>Is +.>；
S5.3, wavelet mother functionPerforming extension and translation, setting->Is>Translation factor->The function after the expansion and the translation is made to be +.>The following steps are:
（4）；
in the method, in the process of the invention,to rely on continuous variable parameter->And->Is a wavelet basis function of (1);
s5.4 for signalsThe following steps are:
（5）；
（6）；
wherein,a space representing all square integrable functions; />For normalizing constant, ++>Is a wavelet transform coefficient;
according to the fourier transform product theorem, equation (6) is written as an equivalent frequency domain expression form:
（7）；
in the method, in the process of the invention,is->Fourier transform of->Is->Harmonic components of different frequencies +.>Plural (i.e. add/drop)>Is the frequency.
6. The method for monitoring the state of an AUV propeller based on pre-detection according to claim 5, wherein the step S5 further comprises the steps of:
s5.5, setting the orthogonal scale function and the wavelet mother function in wavelet packet transformation as respectivelyAnd->The low-pass filter in the filter bank is +.>The high pass filter is +.>According to the theory of multi-resolution analysis, < >>、/>、/>、/>The two-scale equation satisfied between is:
（8）；
in the formula, the coefficient sequenceAnd->The method meets the following conditions:
（9）；
s5.6, introducing a marker:；
definition:
：/> （10）；
for being +.>Generating a general iteration formula of the wavelet packet basis function; />Is a discrete time index, when->When (I)>I.e. the scale function->The method comprises the steps of carrying out a first treatment on the surface of the When->When (I)>I.e. wavelet function；
S5.7, setIndicating signal->In->Layer->The wavelet packet is called wavelet packet coefficient, and the decomposition algorithm of wavelet packet transformation is as follows:
（11）；
in the method, in the process of the invention,；/>；/>；/>low-pass and high-pass filters, respectively, in wavelet decomposition, respectively, with a scaling function +.>And wavelet function->Related to;
the reconstruction algorithm of the wavelet packet transformation is as follows:
（12）。
7. the AUV propeller condition monitoring method based on pre-detection of claim 6, wherein step S6 includes the steps of:
s6.1, constructing an LSTM-MACNN structure, wherein the LSTM structure comprises an LSTM module, three multi-level multi-scale attention convolution modules, namely Section-1, section-2 and Section-3, and a classification module;
s6.2, LSTM Module comprising an input GateThe corresponding weight matrix has ∈ ->The method comprises the steps of carrying out a first treatment on the surface of the Forgetting door->The corresponding weight matrix has ∈ ->The method comprises the steps of carrying out a first treatment on the surface of the An output door->The corresponding weight matrix is；
（13）；
Wherein,for temporary cellular status,/->For the current memory cell state, +.>Is the hidden state of the next cell unit.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410123163.2A CN117647758A (en) | 2024-01-30 | 2024-01-30 | AUV propeller state monitoring system and method based on pre-detection |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410123163.2A CN117647758A (en) | 2024-01-30 | 2024-01-30 | AUV propeller state monitoring system and method based on pre-detection |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117647758A true CN117647758A (en) | 2024-03-05 |
Family
ID=90045497
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410123163.2A Pending CN117647758A (en) | 2024-01-30 | 2024-01-30 | AUV propeller state monitoring system and method based on pre-detection |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117647758A (en) |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109446671A (en) * | 2018-11-01 | 2019-03-08 | 上海海事大学 | A kind of underwater propeller wavelet neural network failure prediction method and device |
| CN110263832A (en) * | 2019-06-11 | 2019-09-20 | 哈尔滨工程大学 | A kind of AUV navigation system method for diagnosing faults based on multiscale analysis |
| CN111275164A (en) * | 2020-01-13 | 2020-06-12 | 浙江大学 | Underwater robot propulsion system fault diagnosis method |
| CN211452846U (en) * | 2019-11-29 | 2020-09-08 | 浙江工业大学 | Fan fault remote monitoring system based on current signal |
| CN112557826A (en) * | 2020-12-08 | 2021-03-26 | 上海海事大学 | Ship electric power system fault diagnosis method |
| CN113283292A (en) * | 2021-04-13 | 2021-08-20 | 江苏大学镇江流体工程装备技术研究院 | Method and device for fault diagnosis of underwater micro propeller |
| CN115437358A (en) * | 2022-10-27 | 2022-12-06 | 中国人民解放军陆军装甲兵学院 | Intelligent state monitoring and fault diagnosis system and fault diagnosis method for industrial robot |
| CN117214591A (en) * | 2023-11-09 | 2023-12-12 | 青岛哈尔滨工程大学创新发展中心 | Fault diagnosis system and method for deep-diving propeller |
-
2024
- 2024-01-30 CN CN202410123163.2A patent/CN117647758A/en active Pending
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109446671A (en) * | 2018-11-01 | 2019-03-08 | 上海海事大学 | A kind of underwater propeller wavelet neural network failure prediction method and device |
| CN110263832A (en) * | 2019-06-11 | 2019-09-20 | 哈尔滨工程大学 | A kind of AUV navigation system method for diagnosing faults based on multiscale analysis |
| CN211452846U (en) * | 2019-11-29 | 2020-09-08 | 浙江工业大学 | Fan fault remote monitoring system based on current signal |
| CN111275164A (en) * | 2020-01-13 | 2020-06-12 | 浙江大学 | Underwater robot propulsion system fault diagnosis method |
| CN112557826A (en) * | 2020-12-08 | 2021-03-26 | 上海海事大学 | Ship electric power system fault diagnosis method |
| CN113283292A (en) * | 2021-04-13 | 2021-08-20 | 江苏大学镇江流体工程装备技术研究院 | Method and device for fault diagnosis of underwater micro propeller |
| CN115437358A (en) * | 2022-10-27 | 2022-12-06 | 中国人民解放军陆军装甲兵学院 | Intelligent state monitoring and fault diagnosis system and fault diagnosis method for industrial robot |
| CN117214591A (en) * | 2023-11-09 | 2023-12-12 | 青岛哈尔滨工程大学创新发展中心 | Fault diagnosis system and method for deep-diving propeller |
Non-Patent Citations (6)
| Title |
|---|
| 于凤芹: "实用小波分析十讲", 30 September 2013, 西安电子科技大学出版社, pages: 103 - 104 * |
| 周文安: "数据科学与大数据技术专业教材丛书 机器学习", 31 August 2021, 北京邮电大学出版社, pages: 184 * |
| 王引苗: "基于深度学习的关系抽取及关系推理", 中国优秀硕士学位论文全文数据库信息科技辑, 15 February 2023 (2023-02-15), pages 17 * |
| 贺家李: "电力系统继电保护原理 增订版", 30 September 2004, 中国电力出版社, pages: 433 * |
| 金花 等: "基于 LSTM-MACNN 的油色谱在线监测装置误告警识别", 电力信息与通信技术, vol. 21, no. 11, 30 November 2023 (2023-11-30), pages 57 * |
| 陈华锋: "基于近端策略优化算法的煤粉锅炉燃烧优化研究", 中国优秀硕士学位论文全文数据库工程科技Ⅱ辑, 15 February 2023 (2023-02-15), pages 14 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111830408B (en) | Motor fault diagnosis system and method based on edge calculation and deep learning | |
| Tang et al. | An improved convolutional neural network with an adaptable learning rate towards multi-signal fault diagnosis of hydraulic piston pump | |
| Zhu et al. | Estimation of bearing remaining useful life based on multiscale convolutional neural network | |
| EP2478423A1 (en) | Supervised fault learning using rule-generated samples for machine condition monitoring | |
| Li et al. | Fault diagnosis method based on encoding time series and convolutional neural network | |
| CN113339204B (en) | Wind driven generator fault identification method based on hybrid neural network | |
| CN116643517B (en) | Accident prevention equipment for underwater manned submersible vehicle | |
| CN115022187B (en) | Situation awareness method and device for electric-gas comprehensive energy system | |
| CN117214591A (en) | Fault diagnosis system and method for deep-diving propeller | |
| CN113553356A (en) | Drilling parameter prediction method and system | |
| CN115758083A (en) | Motor bearing fault diagnosis method based on time domain and time-frequency domain fusion | |
| CN116842379A (en) | Mechanical bearing residual service life prediction method based on DRSN-CS and BiGRU+MLP models | |
| CN111007452A (en) | Fault diagnosis method and device of data acquisition system | |
| Li et al. | Instance weighting-based partial domain adaptation for intelligent fault diagnosis of rotating machinery | |
| CN117647758A (en) | AUV propeller state monitoring system and method based on pre-detection | |
| CN116824479A (en) | Intelligent inspection system for factory | |
| CN116127324A (en) | Fault detection method and device for key components in uninterruptible power supply | |
| CN115826477A (en) | Water area monitoring system and method based on data visualization | |
| Pan et al. | A SENet-TSCNN model developed for fault diagnosis considering squeeze-excitation networks and two-stream feature fusion | |
| Yang et al. | Decision fusion scheme based on mode decomposition and evidence theory for fault diagnosis of drilling process | |
| Zhang et al. | Fault pattern recognition of rolling bearing based on smoothness prior approach and dual-input depth spatial–temporal fusion | |
| CN114118140B (en) | Multi-view intelligent fault diagnosis method and system for explosion-proof motor bearing | |
| CN117033983B (en) | Unmanned ship self-noise detection and identification method and system | |
| CN114279704B (en) | Method and system for monitoring planet-level running state of wind power gear box | |
| Song et al. | Bearing fault diagnosis method using a spatio-temporal neural network based on feature transfer learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |