583 results about "Modal decomposition" patented technology

The invention discloses a vibration fault diagnosis method of a wind turbine generator rolling bearing. The method comprises the following steps of step1, using a vibration sensor to collect wind turbine generator rolling bearing inner ring fault vibration signals; step2, using a model of a bistable system to carry out noise reduction processing on the wind turbine generator rolling bearing inner ring fault vibration signals and outputting a noise reduction signal; step3, carrying out variation modal decomposition on the noise reduction signal output by the step2 so as to acquire IMF components of different frequency ranges; and step4, after the step3, carrying out spectral analysis on each IMF component acquired through VMD decomposition, finding a fault characteristic frequency of a rolling bearing and completing vibration fault diagnosis of the wind turbine generator rolling bearing. By using the vibration fault diagnosis method, accurate diagnosis to a vibration fault of the wind turbine generator rolling bearing can be realized.

The invention provides an ultrasound signal de-noising method based on correlation analysis and empirical mode decomposition (EMD). The method comprises the following steps: firstly fitting the energy curve of the pure noise under ultrasound empty acquisition state to obtain a trend curve Alpha of the noise which changes along with time gain compensation (TGC); then dividing the normally acquired ultrasound echo signals with two continuous frames into two parts, one of which is called main noise signal, the dot product of the echo signals and the Alpha curve and the other of which is called main useful signal, the difference of the echo signals and the first part; carrying out EMD on the main noise part to obtain the energy ratio of the first intrinsic mode function (IMF) component as the weighting coefficient of the signals with two frames and carrying out threshold processing after carrying out correlation analysis on the corresponding IMF component; and finally obtaining the de-noised signal after carrying out weighting reconstruction on the main noise part and the main signal part. The invention has adaptive signal decomposition and noise reduction capabilities, greatly improves the signal to noise ratio of the ultrasound signal and obtains good de-noising effect.

The invention relates to a self-adapting method for filtering vehicle dynamic shaft load signal, aiming to effectively eliminate weighing error due to vehicle dynamic load and remarkably improve the vehicle weighing accuracy. The invention processes the original shaft load signal by zero phase offset and low-pass filtering, so as to eliminate white noise interference and inhibit phase distortion of the shaft load signal during low-pass filtering process. The invention also adopts optimized cut-off to the original weighing signal by searching average value of extreme points of shaft load signal, so as to better inhibit gravitation impulse interference in the weighing signal. Further, the invention decomposes the weighing signal by empirical mode decomposition method, meanwhile, adopts end point continuation method in empirical mode decomposition, so as to eliminate end effect, effectively inhibit the generation of false mode and make the decomposed result relatively true. The invention eliminates low-frequency interference through empirical mode decomposition, so as to obtain true shaft load signal.

The invention discloses a flutter online monitoring method for machining equipment. The method comprises the steps that a proper sampling window is selected; empirical mode decomposition is carried out on sampled vibration signals; decomposed eigen modalities are screened to obtain a characteristic eigen modality; Hilbert transformation is carried out on the characteristic eigen modality to obtain a time-frequency spectrum; statistical pattern analysis is carried out on the time-frequency spectrum to obtain characteristic parameters; the statistical characteristic parameters are compared with a set characteristic threshold value and the statistical characteristic parameter of a historical adjacent signal, and the vibration state of a system is judged. The flutter online monitoring method aims to solve the problems that a flutter detecting method is strong in sample dependency and poor in generalization ability, threshold value measurement is difficult, and judgment is not carried out in time, the method combining Hilbert-Huang transformation and statistical pattern recognition is provided, statistical modeling and clustering analysis are carried out on the time-frequency spectrum of the vibration signal based on the aggregation character of energy on frequency in the fluttering process, the characteristic parameters are utilized, the physical characteristic of cutting flutter is represented essentially, the cutting vibration state is effectively monitored in real time, and the judgment result is accurate and visual.

A visual question and answer method based on a multi-modal decomposition model is provided. The method is characterized in that the image is trained in an ImageNet dataset and image features are extracted, the question is marked as texts into feature vectors; a collaborative attention model is introduced into the basic network architecture, image and question related features are learned, and fine-grained correlation among multi-modal features are characterized; and multi-modal features enter a multi-modal decomposition bilinear pool (MFB) or a multi-modal decomposition high order cell (MFH) module image to generate a fusion image question feature z, and the z is put into the classifier to predict the best matching answer. According to the method provided by the present invention, the collaborative attention model is used to predict the correlation between each spatial grid in the image and the question, and the best matching answer can be accurately predicted in a facilitated manner;and in combination with the image attention mechanism, the model can effectively understand which image region is important to the question, so that the performance of the model and the accuracy of the question and answer can be significantly improved.

The invention belongs to the technical field of methods or devices applying electronic equipment for recognition, and discloses a target radiation source individual recognition method. The target radiation source individual recognition method comprises the steps that a received target radiation source signal is subjected to empirical modal decomposition and variational modal decomposition to obtain different modal components; each intrinsic modal component is denoised through singular spectral analysis, and multi-scale arrangement entropy of each modal component is calculated to construct feature vectors; and the feature vectors are subjected to dimensionality reduction through principal component analysis, and target radiation source individual recognition is achieved through a support vector machine classifier. Target radiation source individual recognition under an environment with the lower signal-to-noise ratio can be effectively achieved, and pretreatment such as signal-to-noiseratio estimation is not needed.

The invention discloses an aquaculture water quality short-time combination forecast method on the basis of multi-scale analysis. The method includes the steps that water quality time sequence data are acquired online and repaired; through empirical mode decomposition, the selected water quality time sequence sample set data are decomposed into IMF components and residual rn components, wherein the IMF components and the residual rn components are different in frequency scale; the IMF components and the rn components are classified, a manual bee colony optimization least square support vector regression machine, a BP neural network and an autoregressive sliding average model are respectively selected for forecast according to classifying features, and finally, all results are weighed and summed to obtain a water quality time sequence forecast result. According to the method, the original water quality time sequence data are decomposed into the components different in time frequency through the empirical mode decomposition, and change conditions in original water quality sequences can be mastered more accurately; advantages of the manual bee colony optimization least square support vector regression machine, advantages of the BP neural network and advantages of the autoregressive sliding average model are complemented and combined, and thus performance of a combined forecast model is effectively improved.

The invention provides a heartbeat frequency detection algorithm of a non-contact vital sign detection system; the steps are: respectively doing bandpass filtering for I and Q two way signals outputted by a continuous wave doppler radar; using a center estimate algorithm to carry out useful dc component recovery; using a complete cluster experience modal decomposition algorithm to separate a heartbeat signal from a human body jitter signal, a breathing signal and environment interference noises; resolving a Hilbert marginal spectrum of each decomposition signal and doing peak value detection; finding out the Hilbert marginal spectrum corresponding to the heartbeat signal according to a marginal spectrum peak position and energy concentration degree close to the spectrum peak; obtaining heartbeat frequency information according to the spectrum peak position. The heartbeat frequency detection algorithm can effectively extract the heartbeat signal under unstable human body and large environment interference noises, thus obtaining accurate heartbeat frequency information, improving anti-interference property of the non-contact vital sign detection system, and satisfying heartbeat frequency detection accuracy demands of medical affairs personnel.

The invention belongs to the technical field of the low-frequency oscillation on-line detection and analysis in an electric power system and provides a low-frequency oscillation modal analysis method, which is based on wide area phasor metrical information and an empirical mode decomposition method and can carry out the on-line detection and node homology grouping toward the complex low-frequency oscillation of the electric power system. The modal analysis method not only has strong adaptability to the complex waveforms of non-linearity, non-stationarity or those containing non-periodic constituents, but also can carry out phase comparison toward nonsine or cosinoidal inherent modal curves which belong to the oscillation mode of the same electric power system but are slightly different in frequency and realize the homology grouping of the nodes participating in each oscillation mode, thus obtaining the power conversion relation among the nodes and the position of an oscillation center or interface.

The invention discloses a short-time traffic-flow combination prediction method. The method comprises the following steps of step 1, presetting an acquisition period and collecting traffic flow time sequence data of a prediction target point according to the acquisition period; if a prediction day of the prediction target point is not a holiday, collecting traffic flow time sequence data of days in at least adjacent previous three weeks, wherein the days are the same weekday with the prediction day; if the prediction day of the prediction target point is a holiday, collecting traffic flow time sequence data of days in at least adjacent previous three years, wherein the days are the same holiday with the prediction day; step 2, using a set experience modal decomposition method to decompose the traffic flow time sequence data into several intrinsic mode components possessing a same characteristic; step 3, using a BP neural network algorithm to carry out prediction on each intrinsic mode component obtained through decomposition by using the set experience modal decomposition method respectively, superposing prediction results of the intrinsic mode components and acquiring a final prediction result. By using the prediction method in the invention, prediction precision of the short-time traffic flow can be effectively increased.

The invention relates to a surface electromyogram signal pattern recognition method based on empirical mode decomposition and fractal. In the conventional method, a single fractal theory is adopted, electromyogram signals are subjected to integral singularity evaluation only, and the local singularity features of the signals are not researched. The method comprises the following steps of firstly, acquiring corresponding surface electromyogram signals on related muscle groups; secondly, extracting multilayer intrinsic mode functions of the electromyogram signals by a method for empirical mode decomposition, and extracting generalized dimension spectrums on each layer of intrinsic mode functions by a method for multifractal analysis; and finally, performing classification and recognition of multiple movement patterns on a classifier by a support vector machine by taking the generalized dimension spectrums on each layer of mode functions as feature vectors for pattern recognition. According to the method, the generalized dimension spectrums on each layer of intrinsic mode functions are extracted as the features of the surface electromyogram signals by the method for multifractal analysis, the robustness is higher, and stable feature data can be calculated in the electromyogram signals at lower signal-to-noise ratio.

The invention discloses a pipeline leakage detection device based on a particle swarm optimization-variation modal decomposition (PSO-VMD) algorithm and a detection method. Sound emission signals without leakage and with leakage are acquired through a sound emission system; firstly, a preset scale K and a penalty coefficient a for sound emission signal decomposing are obtained through the PSO-VMDalgorithm, variation modal decomposition VMD is performed on the signals, and K intrinsic mode function IMF components are obtained; signal reconstruction is performed by adopting a method based on anenergy value, and an optimal observation signal after denoising is obtained; and a domain characteristic parameter of a time domain of a reconstructed signal is taken, and finally leakage detection is performed through a support vector machine (SVM) algorithm. It is achieved that pipeline leakage after happening can be found in time, and the problem that the mistaken alarming rate of a pipeline is high is solved.

The invention discloses a multi-person through-wall time varying breathing signal detection method based on VMD, comprising the steps of: an ultra wide band emission antenna emitting narrow pulse, through human thoracic cavity micro-doppler vibration, echo being received by a receiving antenna, and performing slow time sampling on ultra wide band radar echo to obtain a through-wall human body echo signal matrix; calculating the variance of each distance door through a distance door selection algorithm, the one with a largest variance being a distance door where a multi-person target exists, and employing a low pass filter to eliminate high frequency interference and superfluous frequency components; utilizing a VMD algorithm to perform mode decomposition on filtered signals, and iterating sub-signals to obtain effective information meeting the number and frequency band of breathing; and performing Hilbert transformation and time frequency treatment to obtain dynamic instantaneous information including smooth breathing characteristics. The method can effectively eliminate interference harmonic wave of different breathing components, remove metope interference, and enhance weak breathing signals, and has the characteristics of strong interference immunity and accurate time varying tracking characteristics.

The invention relates to an MRS (magnetic resonance sounding) FID (frequency identity) signal noise inhibition method. The method comprises steps as follows: a signal detected by an MRS system is subjected to spectral analysis, the detected signal is decomposed into a homonymous component X and a quadrature component Y with a normalization quadrature detection technology, and a low-frequency FID signal is obtained through hardware filtering processing; peak noises of the components X and Y in the FID signal are rejected from acquired data respectively with a non-linear energy operator algorithm; preliminary signal and noise separation is performed on the components X and Y respectively on the basis of a PCA (principal component analysis) method; the components X and Y processed with the PCA method are further decomposed on the basis of an EMD (empirical mode decomposition) method, and a signal trend term is extracted; the components X and Y processed with the EMD method are superimposed and averaged, and an e index curve is obtained. The problems including high probability of loss of signal components with a conventional filtering means and the like are completely solved, and various complicated noises included in the MRS FID signal are effectively inhibited.

A modal intervals-based high-speed train bogie fault diagnosis method includes the steps of 1, acquiring a high-speed train bogie vibration signal; 2, preprocessing the high-speed train bogie vibration signal; 3, subjecting the high-speed train bogie vibration signal to modal interval valuing; 4, subjecting the high-speed train bogie vibration signal to EMD (experience modal decomposition); 5, subjecting the high-speed train bogie vibration signal to feature extraction; 6, constructing and training an initial interval neural network model; and 7, performing high-speed train bogie fault diagnosis. Uncertainty in bogie vibration signals is processed by the modal interval method, and high-speed train bogie faults are diagnosed by a classified recognition method of the interval neural network model; the uncertainty in the high-speed train bogie fault diagnosis is handled by the modal interval method, and reliability of fault diagnosis results is greatly improved. The method is applicable to the fault diagnosis for high-speed train bogies.

The invention relates to a multi-group image classification method based on two-dimensional empirical modal decomposition and wavelet denoising, belonging to the filed of image processing. The invention aims at solving the problems of insufficient utilization of image essential characteristics and low classification precision of the traditional classification method. The method comprises the following steps of: firstly, respectively carrying out two-dimensional empirical modal decomposition on each wave band in multi-group images to obtain the former K two-dimensional components and one residual error; secondly, summarizing the former K two-dimensional components as a characteristic value, and obtaining a denoised characteristic value after wavelet denoising; thirdly, randomly and proportionally selecting the denoised characteristic values of a plurality of multi-group images as training samples and test samples of a support vector machine, carrying out parameter training of the support vector machine on the training samples, and then carrying out attribution judgment to form a plurality of sub-classifiers of the support vector machine; and fourthly, constructing multiple classifiers based on a one-to-one strategy by utilizing the sub-classifiers of the support vector machine, and determining the attribution classes of the test samples according to a strategy function to complete the classification of the multi-group images.

The invention discloses a milling flutter recognition method based on variation modal decomposition and energy entropy, and belongs to the technical field of machine tool machining flutter recognition. The milling flutter recognition method comprises the following steps of S1, establishing a VMD mathematical model; S2, establishing a mathematical model of the energy entropy; S3, conducting three sets of milling machining experiments representing three cutting states, including stable cutting, a weak flutter and a severe flutter, and obtaining three sets of milling force signals through a dynamometer; S4, conducting FFT analysis on the three sets of milling force signals, and proving that the three sets of milling force signals represent the states of stable cutting, the weak flutter and the severe flutter respectively; S5, determining the number K of optimal modals of VMD decomposition and a penalty factor alpha through a VMD parameter automatic selection method based on a kurtosis value; S6, solving instantaneous frequencies of multiple IMF, and determining a milling flutter characteristic frequency band; S7, adopting a hammering experiment to obtain the modal of a knife; S8, extracting a flutter characteristic vector of each IMF based on the energy entropy. According to the milling flutter recognition method based on the variation modal decomposition and the energy entropy, the VDM decomposition effect is improved, and automatic flutter recognition is achieved.