A device detection system and method based on fan maintenance

Abstract

The invention discloses an equipment detection system and method based on fan maintenance. The system main program module is a carrier module of system functions, which realizes data import and result export; the time domain analysis module realizes the extraction of signal characteristic parameters; the frequency domain analysis module converts the time domain signal into the frequency domain signal by using the stft function; the signal denoising module denoises the frequency domain signal data by using the VMD algorithm; the feature extraction module realizes feature extraction of the denoised signal data; the fault diagnosis module makes fault judgment according to the range of each feature quantity extracted by the feature extraction, and the method realizes the system process. The invention monitors and analyzes the condition of fan equipment on a computer and no longer relies on expensive equipment; the system has excellent performance, obtains data analysis results in a relatively short time, and generates monitoring results; the system continuously optimizes functions, improves maintenance efficiency, and realizes remote fault diagnosis.

Description

Technical Field

[0001] The present invention relates to the field of fan maintenance, and in particular to an equipment detection system and method based on fan maintenance. Background Art

[0002] In the prior art, wind turbine maintenance mainly relies on fault repair and regular maintenance, and cannot predict the status of components or equipment in advance, causing inconvenience to staff and increasing time and labor costs for actual work.

[0003] Condition-based maintenance is the future application direction. The maintenance method is to analyze the signal. Traditional signal analysis methods mainly rely on specific instruments. These instruments are expensive and have low accuracy, which makes traditional signal analysis methods unable to be widely used. In addition, because the equipment and components inside the fan affect each other and become noise, the signal amplitude and frequency change. When analyzing the equipment status to determine the fault type, the results obtained are not accurate, and it is currently difficult to apply.

[0004] The existing signal analysis methods are mainly divided into two methods: matching based on existing statistical results and denoising signal data:

[0005] The first method is to conduct long-term data monitoring of the equipment, summarize the monitored data results, generate a database, match the newly collected signal with the existing database, obtain a matching result, and determine the type of fault and the degree of fault based on the matching degree.

[0006] The second method is implemented by the signal feature analysis method, which uses the time-frequency analysis method to extract features from the equipment signal and uses the feature quantity results to determine the fault condition. Due to the large amount of noise in the field signal, the feature quantity extraction of the collected signal is inaccurate, which makes it impossible to accurately determine the fault type. Summary of the invention

[0007] The purpose of the present invention is to propose an equipment detection system and method based on fan maintenance in view of the above problems.

[0008] An equipment detection method based on fan maintenance, the steps are:

[0009] S1: Import data: obtain the recorded data of the fan equipment, and the system background automatically converts and stores the data type;

[0010] S2: Time-frequency analysis: extract signal data and understand the time domain characteristics of the signal;

[0011] S3: Frequency domain analysis: The analysis module observes the frequency domain and time-frequency domain results, and the staff manually diagnoses whether there is a fault in the characteristic quantity results based on experience;

[0012] S4: Signal denoising: Using the VMD method optimized by genetic algorithm, the Hurst index is used as the fitness function according to the data characteristics to obtain the optimal number of decomposition layers and penalty factors, and the signal is adaptively denoised to filter out high-frequency noise in the signal to obtain the denoised result;

[0013] S5: Feature extraction: extract features from the denoised signal and display the extraction results in a table for export;

[0014] S6: Fault diagnosis: Fault judgment is performed based on the characteristic value results and each characteristic value range:

[0015] If a certain data is not within the valid range, an error will be reported, the cause will be found, and the fault type will be determined;

[0016] If a certain data is within the valid range, no error will be reported.

[0017] Furthermore, in a device detection method based on fan maintenance, step S2 includes the following sub-steps:

[0018] S21: dividing the signal characteristic parameters into dimensioned parameters and dimensionless parameters;

[0019] S22: Obtaining time domain characteristic information of dimensioned parameters;

[0020] The S22 comprises the following sub-steps:

[0021] S221: Obtain the time domain feature information of the average value, and the extraction formula is:

[0022]

[0023] Among them, X mean , μ represents the signal mean, N represents the signal length, X i Representing signal; S222: obtaining time domain characteristic information of peak-to-peak value;

[0024] S223: Obtain time domain feature information of the root mean square value, and the extraction formula is:

[0025]

[0026] Among them, X rms represents the RMS value of the signal;

[0027] S224: Obtain time domain feature information of the variance, and the extraction formula is:

[0028]

[0029] Among them, X var represents the signal variance;

[0030] S225: Obtain time domain feature information of the standard deviation, and the extraction formula is:

[0031]

[0032] Among them, X std ,σ represents the signal standard deviation;

[0033] S23: Obtaining time domain characteristic information of dimensionless parameters;

[0034] The S23 comprises the following sub-steps:

[0035] S231: Obtain the time domain feature information of the skewness, and the extraction formula is:

[0036]

[0037] Among them, skew represents the signal skewness, and E represents the mathematical expectation;

[0038] S232: Obtain the time domain feature information of the kurtosis, and the extraction formula is:

[0039]

[0040] Where ku represents the signal kurtosis;

[0041] S233: Obtain time domain characteristic information of the waveform factor, and the extraction formula is:

[0042]

[0043] Where S represents the signal waveform factor;

[0044] S234: Obtain time domain characteristic information of the margin factor, and the extraction formula is:

[0045]

[0046] Where L represents the signal margin factor;

[0047] S235: Obtain the time domain characteristic information of the pulse factor, and the extraction formula is:

[0048]

[0049] Where I represents the impulse factor.

[0050] Further, in an equipment detection method based on fan maintenance, the dimensioned parameters include mean value, peak-to-peak value, root mean square value, variance and standard deviation;

[0051] The dimensionless parameters include skewness, kurtosis, waveform factor, margin factor, and impulse factor;

[0052] The average value is the mean value of the signal in the acquisition time period;

[0053] The peak-to-peak value is the difference between the signal peak value and the signal trough value;

[0054] The variance is the degree of change of the amplitude of the measured signal around the mean value during the sampling period;

[0055] The standard deviation is the square root of the signal variance value, reflecting the change in signal energy;

[0056] The skewness indicates the symmetry of the signal data distribution and reflects the degree of skewness of the signal;

[0057] The kurtosis is the degree of flatness of the signal waveform;

[0058] The margin factor is the ratio of the signal peak value to the square root amplitude;

[0059] The pulse factor is the ratio of the signal peak value to the rectified average value, and is used to detect whether there is an impact signal in the signal.

[0060] Further, in a device detection method based on fan maintenance, step S3 includes the following sub-steps:

[0061] S31: Assume that a period is T, obtain a periodic signal, and the formula of the periodic signal is:

[0062]

[0063] Among them, x(t) represents a periodic signal; a0, a n 、b n represents the Fourier coefficient; f n represents the frequency of each harmonic; T represents the cycle time; t represents time;

[0064] S32: Set the period of the transient signal to infinity, and perform Fourier transform on the periodic signal. The formula for Fourier transform of the periodic signal is:

[0065]

[0066] Where y(f) represents the Fourier transform of x(t); where j represents the imaginary unit;

[0067] S33: Perform A / D conversion on the periodic signal to convert it into a signal sequence, and obtain the Fourier transform of the discrete signal. The formula of the Fourier transform of the discrete signal is:

[0068]

[0069] Where y(k) represents the Fourier transform of the discrete signal; n and k represent the sequence numbers; x(n) represents the discrete signal sequence; N represents the signal length;

[0070] S34: When the A / D conversion rate is fs and the number of signal sampling points is an even number, fs / N is the signal frequency at the sequence number k=N / 2;

[0071] S35: When the number of signal sampling points is an integer power of 2, the FFT algorithm with a base of 2 is used for calculation, and the resulting complex number and complex modulus value are the amplitude value of the signal at the frequency, and the amplitude values ​​of all points in the signal sequence constitute the amplitude spectrum of the signal;

[0072] S36: The ratio of the imaginary part and the real part of the number obtains the phase value of the signal at the frequency. The phase values ​​of all points in the signal sequence constitute the phase spectrum of the signal. The total energy accumulated in the time domain of the signal is equal to the total energy accumulated in the frequency domain by the Fourier transform of the signal. The Fourier transform formula of the periodic signal is transformed into the time domain form:

[0073]

[0074] The formula for the Fourier transform of a discrete signal is transformed into the time domain form:

[0075]

[0076] S37: Perform Fourier transform on the collected signal, square the modulus value of the calculation result, and then divide it by the signal length to obtain the discrete signal power. The formula is:

[0077]

[0078] Among them, P^ PER (k) represents the discrete signal power value.

[0079] Furthermore, an equipment detection method based on fan maintenance is characterized in that the sub-steps of step S4 include:

[0080] S41: Decompose the signal by using variational mode decomposition, and preset the decomposition parameters of the decomposition layer number and the penalty factor;

[0081] S42: calculating the Hurst index of the decomposed signal, and adjusting the preset parameters according to the size of the Hurst index;

[0082] S43: Use the genetic algorithm to take the Hurst index as the fitness value, iterate and evolve, optimize the preset parameters, and obtain the most effective decomposition result.

[0083] Further, in a device detection method based on fan maintenance, step S41 includes the following sub-steps:

[0084] S411: Assume that the original signal s is divided into k components v, each mode has a center frequency, and the sum of the estimated bandwidths of each mode is minimized. The sum of all modes is equal to the original signal. The constrained variational expression is:

[0085]

[0086] where {v k} = {v1, …, v k} are the decomposed IMF components; {ω k} = {ω1, …, ω k} are the center frequencies of each component; k is the number of components; s is the original signal; is to take the partial derivative; δ(t) is the Dirac distribution function;

[0087] S412: Introduce the Lagrange multiplier τ(t) and the second-order penalty factor α to transform the constrained variational problem into an unconstrained variational problem, and obtain a more general Lagrange multiplier expression:

[0088]

[0089] where w represents the angular frequency;

[0090] S413: Use the alternating direction multiplier method to solve the unconstrained variational problem. During the iterative solution process, continuously update and until the iteration stop condition is satisfied, and complete the solution of the unconstrained variational problem.

[0091] Furthermore, a device detection method based on fan maintenance, the step S42 includes the following sub-steps:

[0092] S421: Use the variational mode decomposition method to decompose the original signal, calculate the eigenvalue of the decomposed signal, and the eigenvalue is used as the Hurst exponent H;

[0093] The Hurst exponent H has three forms, including normal, low, and high:

[0094] Normal: H = 0.5, indicating that the time series can be described by a random walk and the trend cannot be judged;

[0095] Low: 0.5 < H < 1, indicating that the time series has long-term memory and there is a positive correlation between the past and the future. The closer H is to 1, the stronger the positive correlation and the stronger the trend of the series;

[0096] High: 0 ≤ H < 0.5, indicating the pink noise mean reversion process. The closer H is to 0, the stronger the negative correlation and the stronger the oscillation mutation;

[0097] S422: Make a judgment on the Hurst exponent H:

[0098] If the Hurst index is high, it means that the information in the modal component is mixed and there is a lot of noise;

[0099] If the Hurst index is low, it means that the information in the modal component is stable and there is little noise;

[0100] S423: Using the Hurst exponent, several modal components similar to the original signal are found, and each modal component is reconstructed to achieve signal denoising.

[0101] Further, in a device detection method based on fan maintenance, step S43 includes the following sub-steps:

[0102] S431: Using genetic algorithms to preset parameters for encoding, initializing genomes, evaluating the adaptability, selection, crossover and mutation of individuals in the genome, optimizing the effect, using the Hurst index as the fitness value, constantly adjusting the parameter values, and gradually optimizing using the rule of survival of the fittest;

[0103] The population is two parameters with upper and lower limits;

[0104] S432: Iterate to generate approximate solutions, continuously iterate and evolve, and finally obtain the optimal solution of the decomposition result.

[0105] An equipment detection system based on fan maintenance includes a main program module, a time domain analysis module, a frequency domain analysis module, a signal denoising module, a feature extraction module, and a fault diagnosis module;

[0106] The main program module is the carrier module of the system functions, which realizes the import of data and the export of results;

[0107] The time domain analysis module realizes the extraction of signal characteristic parameters and understands the time domain characteristics of the signal;

[0108] The frequency domain analysis module uses the stft function to convert the time domain signal into the frequency domain signal;

[0109] The signal denoising module uses the VMD algorithm to denoise the frequency domain signal data;

[0110] The feature extraction module realizes feature extraction of the denoised signal data;

[0111] The fault diagnosis module makes fault judgments based on the range of each feature quantity extracted, and issues real-time alarms when an abnormality is about to occur or has occurred.

[0112] The beneficial effects of the present invention are as follows: through an equipment detection system and method based on fan maintenance, a series of difficulties in the prior art, such as expensive dedicated acquisition and analysis equipment, signal feature quantities containing more noise, difficulty in signal extraction, too much acquisition data, difficulty in analysis, poor portability, and poor scalability, are solved, and fan equipment condition monitoring and analysis on a computer is realized, which no longer relies on expensive equipment and has a low cost; the system has excellent performance and can obtain data analysis results and generate monitoring results in a relatively short time; the system has functions such as signal denoising, signal analysis, and feature quantity analysis; the system is less affected by hardware, the quality can be improved, the function can be continuously optimized, the maintenance efficiency can be improved, and remote fault diagnosis can be realized. BRIEF DESCRIPTION OF THE DRAWINGS

[0113] Figure 1 It is a system structure diagram of the present invention.

[0114] Figure 2 It is a flow chart of the present invention.

[0115] Figure 3 Flowchart of the genetic algorithm of the present invention. DETAILED DESCRIPTION

[0116] In order to have a clearer understanding of the technical features, purposes and effects of the present invention, specific embodiments of the present invention are now described with reference to the accompanying drawings.

[0117] As attached Figure 1 As shown, the module structure is:

[0118] It includes main program module, time domain analysis module, frequency domain analysis module, signal denoising module, feature extraction module and fault diagnosis module;

[0119] 1.1 Main program module:

[0120] The module that carries the main functions of the system. After logging into the software and importing the data, the collected signals can be displayed on the software. Select the corresponding function to perform simple time-frequency domain analysis and extract characteristic quantities such as peak value, variance and standard deviation. The extraction results will be displayed in a table and can be exported.

[0121] 1.2 Time domain analysis module:

[0122] The signal time domain characteristic parameters represent the time domain characteristic information of the signal. By extracting the characteristic parameters of the measured signal, the time domain characteristics of the signal can be understood.

[0123] Signal characteristic parameters are divided into two categories: one is dimensional, including parameters such as mean value, peak-to-peak value, root mean square value, variance and standard deviation; the other is a series of dimensionless parameters, including skewness, kurtosis, waveform factor, margin factor, pulse factor and other parameters.

[0124] The parameters and corresponding formulas are as follows:

[0125] (1) Average value: the average value of the signal during the acquisition period. The formula is:

[0126]

[0127] Where X mean , μ represents the signal mean, N represents the signal length, X i Represents a signal.

[0128] (2) Peak-to-peak value: The difference between the value corresponding to the signal peak and the value corresponding to the signal trough.

[0129] (3) The formula for the root mean square value is:

[0130]

[0131] Where X rms is the root mean square value.

[0132] (4) Variance: The degree of variation of the measured signal amplitude around the mean value during the sampling period. The formula is:

[0133]

[0134] Where X var is the signal variance.

[0135] (5) Standard deviation: The standard deviation is the square root of the signal variance value, which reflects the change in signal energy. The formula is:

[0136]

[0137] Where X std and σ is the signal standard deviation.

[0138] (6) Skewness: Signal skewness indicates the symmetry of the signal data distribution and reflects the degree of skewness of the signal. The following is the signal skewness formula:

[0139]

[0140] Where skew is the signal skewness and E is the mathematical expectation.

[0141] (7) Kurtosis: Kurtosis describes the degree of smoothness of the signal waveform. Its formula is:

[0142]

[0143] Where ku is the signal kurtosis.

[0144] (10) Waveform factor: The ratio of the effective value of the signal to the root mean square amplitude. The formula is:

[0145]

[0146] Where S is the signal waveform factor.

[0147] (11) Margin factor: The margin factor is the ratio of the signal peak value to the square root amplitude. Its formula is:

[0148]

[0149] Where L is the signal margin factor.

[0150] (12) Pulse factor: The pulse factor is the ratio of the signal peak value to the rectified average value (average of the absolute value). It is used to detect whether there is an impulse signal in the signal. The pulse factor formula is:

[0151]

[0152] Where I is the pulse factor.

[0153] 1.3 Frequency domain analysis module:

[0154] According to Fourier transform theory, all signals can be formed by combining multiple sinusoidal signals at a specific frequency. Suppose there is a signal x(t) with a period of T, that is:

[0155]

[0156] Where x(t) represents a periodic signal; a0, a n 、b n represents the Fourier coefficient; f n represents the frequency of each harmonic; T represents the cycle time; t represents time.

[0157] For transient signals, their period can be considered infinite, and the Fourier transform of the signal x(t) is:

[0158]

[0159] Where y(f) is the Fourier transform of x(t); j is the imaginary unit.

[0160] The signal is converted into a signal sequence x(n) by A / D conversion. Then the Fourier transform (DFT) of the discrete signal x(n) is:

[0161]

[0162] Where n and k are serial numbers; x(n) is the discrete signal sequence; N is the signal length; and y(k) is the Fourier transform of the discrete signal.

[0163] If the A / D conversion rate is fs and the number of signal sampling points is an even number, then fs / N is the signal frequency at the serial number k=N / 2. When the number of signal sampling points is an integer power of 2, the FFT algorithm with a base of 2 can be used for calculation. The result is a complex number, and the complex modulus is the amplitude value of the signal at the frequency. The amplitude values ​​of all points in the signal sequence constitute the amplitude spectrum of the signal.

[0164] The ratio of the imaginary part and the real part of the complex number gives the phase value of the signal at that frequency. The phase values ​​of all points in the signal sequence constitute the phase spectrum of the signal. According to Parseval's theorem, the total energy accumulated in the time domain of the signal is equal to the total energy accumulated in the frequency domain by the Fourier transform of the signal. The formula is:

[0165]

[0166] Where x(t) is a periodic signal; |y(f)| is the Fourier transform of the periodic signal.

[0167] For discrete signals:

[0168]

[0169] Where x(n) is the discrete signal; N is the length of the discrete signal; y(k) is the Fourier transform of the discrete signal.

[0170] Therefore, the collected signal is transformed by Fourier, the result is modulo and then squared, and then divided by the signal length N, the discrete signal power value can be obtained:

[0171]

[0172] Where P^PER(k) is the power value of the discrete signal; N is the length of the discrete signal; y(k) is the Fourier transform of the discrete signal.

[0173] 1.4 Time-frequency analysis module:

[0174] The module program is designed using the Python extension tool library Scipy. Scipy is a data processing and analysis toolkit for Python, which has functions such as scientific computing, signal processing, and image processing. Signal is a signal processing toolkit in Scipy, which contains many signal processing functions such as Fourier transform function, short-time Fourier transform function, wavelet transform function, and Hilport transform. The stft function in the Signal module is used to complete the short-time Fourier transform of the collected signal, and the Pytftb package is used for function expansion to complete richer time-frequency analysis functions.

[0175] 1.5 Signal Denoising Module

[0176] After importing the data, the software converts the data type internally to facilitate data extraction and analysis. The software's VMD adaptive denoising function uses an improved VMD algorithm, which can adaptively denoise the existing signal data. After removing irrelevant noise, the denoised signal result can be obtained, and the data can be saved and analyzed as needed.

[0177] Variational mode decomposition (VMD) is an adaptive, non-recursive mode variation and signal processing technology that can continuously optimize the mode function and center frequency through iterative search to obtain several mode functions with a certain bandwidth, thereby better meeting the needs of complex signal processing.

[0178] Compared with the EMD method, the VMD method has a stronger mathematical theoretical basis, can decompose non-stationary signals more accurately, is less prone to modal aliasing, and has good noise resistance and efficient computing capabilities.

[0179] Scholars at home and abroad have successfully applied VMD to various signals such as radar signals and speech signals, and have achieved certain results in the field of signal denoising.

[0180] Assume that the original signal s is divided into k components v, each mode has a center frequency, and the sum of the estimated bandwidths of each mode is the smallest. The sum of all modes is equal to the original signal. The constrained variational expression is:

[0181]

[0182] In the formula {v k}={v1,…,v k} is the decomposed IMF component; {ω k}={ω1,…,ω k} is the center frequency of each component; k is the number of components; s is the original signal; To find the partial derivative; δ(t) is the Dirac distribution function.

[0183] In order to effectively seek the optimal solution to the constrained variational problem, the Lagrange multiplier τ(t) and the second-order penalty factor α are introduced to transform the constrained variational problem into an unconstrained variational problem, thereby obtaining a more extensive Lagrange expression:

[0184]

[0185] Where w represents the angular frequency.

[0186] The alternating direction multiplier method (DMM) can effectively solve the unconstrained variational problem. In the iterative solution process, the sum is continuously updated until the iteration stop condition is met, and finally the unconstrained variational problem is solved.

[0187] The Hurst exponent was first proposed by the British scientist Hurst and can help us identify whether a time series is affected by time. Initially, it could be used to measure the flow of reservoirs and rivers. Later, it was also applied to the fractal analysis of various industries to more accurately reflect the actual situation.

[0188] The Hurst exponent has three forms:

[0189] (1) If H = 0.5, it indicates that the time series can be described by a random walk and the trend cannot be judged.

[0190] (2) If 0.5 < H < 1, it indicates that the time series has long-term memory and there is a positive correlation between the past and the future. The closer H is to 1, the stronger the positive correlation and the stronger the trend of the series.

[0191] (3) If 0 ≤ H < 0.5, it indicates pink noise (anti-persistence), that is, a mean-reverting process. The closer H is to 0, the stronger the negative correlation and the stronger the oscillatory mutation.

[0192] After VMD decomposition, when the Hurst exponent is low, it means that the information in the modal components is mixed and there is more noise; while when the Hurst exponent is high, it means that the information in the modal components is more stable and there is less noise. The Hurst exponent is used to find several modal components that are most similar to the original signal, and then the modal components are reconstructed to finally achieve signal denoising.

[0193] When using the VMD method to decompose a signal, decomposition parameters need to be preset. Among them, the decomposition level and the penalty factor are the two parameters that have the greatest impact on the decomposition result. Usually, when performing VMD decomposition, it is mostly based on empirical preset. However, the actual signal is relatively complex, and the effect of empirical preset is not ideal. Moreover, the two parameters affect each other. By fixing one parameter to solve the other parameter, it will fall into the deadlock of local optimization and the best decomposition effect cannot be obtained. Therefore, seeking the optimal decomposition parameters of the decomposition level and the penalty factor is the key to the VMD method.

[0194] As shown in the appendix Figure 3 Using the genetic algorithm, taking the Hurst exponent of the components after VMD decomposition as the fitness value, iteratively generating better and better approximate solutions, continuously iterating and evolving, and finally obtaining the optimal solution. Using the genetic algorithm to solve optimization problems usually includes 6 key steps: encoding, initializing the genome, evaluating the fitness of individuals in the genome, selection, crossover, and mutation, so that the population evolves into a new generation of populations with better fitness.

[0195] 1.6 Feature extraction module:

[0196] The obtained signal data is subjected to feature extraction. Each feature quantity has a normal range. The fault type is judged based on the size of the feature quantity, and a real-time alarm is issued to determine whether an abnormality has occurred or is about to occur.

[0197] 2. Introduction to the overall processing flow and functions of the software:

[0198] 2.1 Process:

[0199] As attached Figure 2 As shown, data import, time-frequency analysis, signal denoising, feature extraction, and fault diagnosis.

[0200] (1) Obtain the recorded data of the fan equipment, and the system background automatically converts and stores the data type.

[0201] (2) Perform simple time-frequency analysis on the signal data, and use the time-frequency analysis module, frequency domain analysis module and time-frequency analysis module as described above to observe the time domain, frequency domain and time-frequency domain results. At this time, manual diagnosis can be performed based on experience.

[0202] (3) Using the VMD method optimized by the genetic algorithm mentioned in the signal denoising module 1.5, the Hurst index is used as the fitness function according to the data characteristics to obtain the optimal number of decomposition layers and penalty factors, and the optimal parameters are used to adaptively denoise the signal, filter out high-frequency noise in the signal, and obtain the denoised result.

[0203] (4) Feature extraction of the denoised signal can obtain more accurate feature quantity results.

[0204] (5) Use the characteristic quantity results to compare with the established range. When a certain data is not within the valid range, an error is reported, the cause is found, and the fault type is determined.

[0205] (6)2.2 Function

[0206] By monitoring and analyzing the equipment data, the denoising method is used to make the obtained data more accurate. By analyzing the data at this time, the characteristic quantity obtained is closer to the real result, and the equipment status can be obtained by comparing it with the normal range. A result similar to a physical examination report can be obtained. According to the result, if there is any abnormality, an alarm can be triggered and handled by the staff. Fault handling is carried out when the equipment abnormality begins.

[0207] This solution, through an equipment detection system and method based on fan maintenance, solves a series of difficulties in the prior art, such as expensive dedicated acquisition and analysis equipment, signal feature quantities containing more noise, difficulty in signal extraction, too much collected data, difficulty in analysis, poor portability, and poor scalability. It realizes the monitoring and analysis of fan equipment conditions on a computer, no longer relying on expensive equipment, and has a low cost; the system has excellent performance, can obtain data analysis results in a relatively short time, and generate monitoring results; the system has functions such as signal denoising, signal analysis, and feature quantity analysis; the system is less affected by hardware, the quality can be improved, and the function can be continuously optimized to improve maintenance efficiency and realize remote fault diagnosis.

[0208] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A device detection method based on fan maintenance, characterized in that: Use the equipment detection system based on fan maintenance for detection, the detection system includes a main program module, a time domain analysis module, a frequency domain analysis module, a signal denoising module, a feature extraction module, and a fault diagnosis module; The main program module is the carrier module of the system functions, which realizes the import of data and the export of results; The time domain analysis module realizes the extraction of signal characteristic parameters and understands the time domain characteristics of the signal; The frequency domain analysis module uses the stft function to convert the time domain signal into the frequency domain signal; The signal denoising module uses the VMD algorithm to denoise the frequency domain signal data; The feature extraction module realizes feature extraction of the denoised signal data; The fault diagnosis module makes fault judgments based on the range of each feature quantity extracted from the features, and issues a real-time alarm when an abnormality is about to occur or has occurred; The steps of the detection method are: S1: Import data: obtain the recorded data of the fan equipment, and the system background automatically converts and stores the data type; S2: Time-frequency analysis: extract signal data and understand the time domain characteristics of the signal; S3: Frequency domain analysis: The analysis module observes the frequency domain and time-frequency domain results, and the staff manually diagnoses whether there is a fault in the characteristic quantity results based on experience; S4: Signal denoising: Using the VMD method optimized by genetic algorithm, the Hurst index is used as the fitness function according to the data characteristics to obtain the optimal number of decomposition layers and penalty factors, and the signal is adaptively denoised to filter out high-frequency noise in the signal to obtain the denoised result; S5: Feature extraction: extract features from the denoised signal and display the extraction results in a table for export; S6: Fault diagnosis: Fault judgment is performed based on the characteristic value results and each characteristic value range: If a certain data is not within the valid range, an error will be reported, the cause will be found, and the fault type will be determined; If a certain data is within the valid range, no error will be reported; The step S3 comprises the following sub-steps: S31: Assume that a period is T, obtain a periodic signal, and the formula of the periodic signal is: Among them, x(t) represents a periodic signal; a0, a n 、b n represents the Fourier coefficient; f n represents the frequency of each harmonic; T represents the cycle time; t represents time; S32: Set the period of the transient signal to infinity, and perform Fourier transform on the periodic signal. The formula for Fourier transform of the periodic signal is: Where y(f) represents the Fourier transform of x(t); where j represents the imaginary unit; S33: Perform A / D conversion on the periodic signal to convert it into a signal sequence, and obtain the Fourier transform of the discrete signal. The formula of the Fourier transform of the discrete signal is: Where y(k) represents the Fourier transform of the discrete signal; n and k represent the sequence numbers; x(n) represents the discrete signal sequence; N represents the signal length; S34: When the A / D conversion rate is fs and the number of signal sampling points is an even number, fs / N is the signal frequency at the sequence number k=N / 2; S35: When the number of signal sampling points is an integer power of 2, the FFT algorithm with a base of 2 is used for calculation, and the resulting complex number and complex modulus value are the amplitude value of the signal at the frequency, and the amplitude values ​​of all points in the signal sequence constitute the amplitude spectrum of the signal; S36: The ratio of the imaginary part and the real part of the number obtains the phase value of the signal at the frequency. The phase values ​​of all points in the signal sequence constitute the phase spectrum of the signal. The total energy accumulated in the time domain of the signal is equal to the total energy accumulated in the frequency domain by the Fourier transform of the signal. The Fourier transform formula of the periodic signal is transformed into the time domain form: The formula for the Fourier transform of a discrete signal is transformed into the time domain form: S37: Perform Fourier transform on the collected signal, square the modulus value of the calculation result, and then divide it by the signal length to obtain the discrete signal power. The formula is: Among them, P^ PER (k) represents the discrete signal power value; The sub-steps of step S4 include: S41: Decompose the signal by using variational mode decomposition, and preset the decomposition parameters of the decomposition layer number and the penalty factor; S42: Calculating the Hurst index of the decomposed signal, and adjusting the preset parameters according to the size of the Hurst index; S43: Use the genetic algorithm to take the Hurst index as the fitness value, iterate and evolve, optimize the preset parameters, and obtain the optimal decomposition result.

2. The equipment detection method based on fan maintenance according to claim 1 is characterized in that: The step S2 comprises the following sub-steps: S21: dividing the signal characteristic parameters into dimensioned parameters and dimensionless parameters; S22: Obtaining time domain characteristic information of dimensioned parameters; The S22 comprises the following sub-steps: S221: Obtain the time domain feature information of the average value, and the extraction formula is: Among them, X mean , μ represents the signal mean, N represents the signal length, X i Represents a signal; S222: Obtaining time domain characteristic information of peak-to-peak value; S223: Obtain time domain feature information of the RMS value, and the extraction formula is: Among them, X rms represents the RMS value of the signal; S224: Obtain time domain feature information of the variance, and the extraction formula is: Among them, X var represents the signal variance; S225: Obtain time domain feature information of the standard deviation, and the extraction formula is: Among them, X std ,σ represents the signal standard deviation; S23: Obtaining time domain characteristic information of dimensionless parameters; The S23 comprises the following sub-steps: S231: Obtain the time domain feature information of the skewness, and the extraction formula is: Among them, skew represents the signal skewness, and E represents the mathematical expectation; S232: Obtain the time domain feature information of the kurtosis, and the extraction formula is: Where ku represents the signal kurtosis; S233: Obtain time domain characteristic information of the waveform factor, and the extraction formula is: Where S represents the signal waveform factor; S234: Obtain time domain characteristic information of the margin factor, and the extraction formula is: Where L represents the signal margin factor; S235: Obtain the time domain characteristic information of the pulse factor, and the extraction formula is: Where I represents the impulse factor.

3. The equipment detection method based on fan maintenance according to claim 2 is characterized in that: The dimensioned parameters include mean value, peak-to-peak value, root mean square value, variance and standard deviation; The dimensionless parameters include skewness, kurtosis, waveform factor, margin factor, and impulse factor; The average value is the mean value of the signal in the acquisition time period; The peak-to-peak value is the difference between the signal peak value and the signal trough value; The variance is the degree of change of the amplitude of the measured signal around the mean value during the sampling period; The standard deviation is the square root of the signal variance value, reflecting the change in signal energy; The skewness indicates the symmetry of the signal data distribution and reflects the degree of skewness of the signal; The kurtosis is the degree of flatness of the signal waveform; The margin factor is the ratio of the signal peak value to the square root amplitude; The pulse factor is the ratio of the signal peak value to the rectified average value, and is used to detect whether there is an impact signal in the signal.

4. According to claim 1, a method for detecting equipment based on fan maintenance is characterized in that: The step S41 includes the following sub-steps: S411: Assume that the original signal s is divided into k components v, each mode has a central frequency, and the sum of the estimated bandwidths of each mode is minimized. The sum of all modes is equal to the original signal. The constrained variational expression is: where, {v k} = {v 1, …, v k} represents the decomposed IMF components; {ωk} = {ω1, …, ωk} represents the central frequencies of each component; k represents the number of components; s represents the original signal; ∂ / ∂ represents taking the partial derivative; δ(t) represents the Dirac distribution function. S412: Introduce the Lagrange multiplier τ(t) and the second-order penalty factor α to transform the constrained variational problem into an unconstrained variational problem, and obtain a more general Lagrange multiplier expression: where, w represents the angular frequency. S413: Use the alternating direction multiplier method to solve the unconstrained variational problem. During the iterative solution process, continuously update and until the iteration stop condition is satisfied, and complete the solution of the unconstrained variational problem.

5. The equipment detection method based on fan maintenance according to claim 1 is characterized in that: The step S42 includes the following sub-steps: S421: Use the variational mode decomposition method to decompose the original signal, calculate the eigenvalues of the decomposed signal, and the eigenvalues are used as the Hurst exponent H. The Hurst exponent H has three forms, including normal, low, and high: Normal: H = 0.5, indicating that the time series can be described by a random walk and the trend cannot be judged. Low: 0.5 < H < 1, indicating that the time series has long-term memory and there is a positive correlation between the past and the future. The closer H is to 1, the stronger the positive correlation and the stronger the trend of the series. High: 0 ≤ H < 0.5, indicating the mean reversion process of pink noise. The closer H is to 0, the stronger the negative correlation and the stronger the oscillation mutation. S422: Make a judgment on the Hurst exponent H: If the Hurst exponent is high, it means that the information in the modal components is mixed and there is a lot of noise. If the Hurst exponent is low, it means that the information in the modal components is stable and there is little noise. S423: Use the Hurst exponent to find several modal components similar to the original signal, and reconstruct each modal component to achieve signal denoising.

6. The equipment detection method based on fan maintenance according to claim 1 is characterized in that: The step S43 includes the following sub-steps: S431: Use the genetic algorithm to preset parameters for encoding, initialize the genome, evaluate the fitness of individuals in the genome, select, crossover, and mutate respectively to optimize the effect. Use the Hurst exponent as the fitness value, continuously adjust the parameter values, and gradually optimize using the rule of survival of the fittest. Use the genetic algorithm to solve the optimization problem and make the population evolve into a new generation of better-adapted population. The population is two parameters with upper and lower limits. S432: Iteratively generate approximate solutions, continuously iterate and evolve, and finally obtain the optimal solution of the decomposition result.

Images