48 results about "Recurrence plot" patented technology

The invention relates to a performance evaluation method for a rolling bearing under multiple operating conditions based on geometric measurement. The performance evaluation method comprises the steps of firstly converting rolling bearing vibration signals with different fault degrees under different operating conditions into a recurrence plot, then stable extracting geometric features of the converted recurrence plot by using scale invariant feature transform method, carrying out singular value decomposition on the geometric features extracted from the bearing signals to acquire feature curves of the rolling bearing with different fault degrees under the different operating conditions, and then calculating the similarity of different feature curves by using dynamic time warping so as to work out a health degree of the rolling bearing. Case verification is carried out by using bearing data of the Case Western Reserve University in the United States, and an experimental result shows that the performance evaluation method can evaluate the health degree of the rolling bearing under multiple operating conditions effectively.

The invention discloses a similarity mining method of similar behaviors between multiple views and a behavior recognition method. The behavior recognition method is based on the similarity mining method. The similarity mining method comprises calculating a Hessian matrix response of every pixel in every frame of gray images of video streaming of different views of the similar behaviors frame by frame; determining feature points according to the matrix responses; building space-time feature descriptors of the feature points and obtaining posture description vectors after dimensionality reduction; calculating a similarity recurrence plot matrix with the posture description vectors as low-level features; and obtaining recursiveness descriptors according to the recurrence plot matrix and clustering the recursiveness descriptors. According to the similarity mining method of the similar behaviors between the multiple views and the behavior recognition method, posture modeling is achieved through a space-time feature point method and recursive similarity between the multiple views is mined through a recurrence plot, so that the problem that background segmentation dependency is easy to cause distortion of human body postures is solved and the accuracy of similarity mining and the robustness of behavior recognition are improved.

A system for measuring heart rate variability (HRV) of a patent comprises: recording means for obtaining and regulating heartbeat-to-heartbeat intervals for a predetermined period of time; processing means for digitizing said intervals, forming a recurrence plot, and assigning a unit mass to each point on the plot representing a measured interval, and calculating the determinant by the expression<paragraph lvl="0"><in-line-formula>Qdet=QxxQyy</in-line-formula>wherein: Qxx is the quadrupole moment relative to the X axis of the principal coordinate, Qyy is the quadrupole moment relative to the Y axis of the principal coordinate; and Qdet is the product of Qxx and Qyy.

The invention provides a multi-target SSVEP idea control method and the application thereof based on the integration of recurrence plots and the deep learning. According to the invention, the phase information is added, and a multi-target SSVEP electroencephalogram experiment stimulation interface is designed. Meanwhile, n SSVEP electroencephalogram signals induced by n stimulating pictures of each tested person in more than eight subjects are obtained. At the same time, the recursion plots of the brain electrical signals of more than eight subjects under the induction of different stimulationpictures are obtained. A label is set for each recursion plot to serve as a sample, and a data set is constructed. After that, a deep convolution neural network model structure and parameters thereofare built and optimized. In this way, a depth convolution neural network model which can be used for effectively classifying the recursion plots of SSVEP electroencephalogram signals induced by different stimulation pictures is determined. A new tested person SSVEP electroencephalogram signal is reconstructed through the phase space, and the optimized depth convolution neural network model is input in the recursion plot form. As a result, the accurate classification of multi-target SSVEP electroencephalogram signals is achieved and an idea control instruction is generated. The multi-objectiveidea control is achieved. The method is suitable for being applied to the field of the complex control of multiple targets.

The invention discloses a rolling bearing variable-work-condition fault diagnosis method based on visual cognition, and relates to a rolling bearing variable-work-condition fault diagnosis technology. The method comprises the following steps of converting rolling bearing vibration signals under the variable work conditions into a two-dimensional image by using a recurrence plot technology; performing feature extraction on the two-dimensional image by utilizing an SURF (speed up robust features) algorithm to obtain the vision invariability high-dimension fault feature vector; performing dimension reduction processing on the high-dimension feature vector by using an equal-distance mapping Isomap algorithm to obtain the low-dimension stable feature vector; using an SVD (singular value decomposition) algorithm for extracting the feature matrix singular value built by the low-dimension stable feature vector to form the final feature vector; performing fault classification on the final feature vector by using the trained classifier; performing fault diagnosis on the rolling bearing under the variable work conditions. The invention provides a novel solution for the rolling bearing fault diagnosis.

The invention discloses a motor bearing fault diagnosis method based on a recurrence plot and a multilayer convolutional neural network. The method comprises the following steps: collecting one-dimensional vibration signals of motor bearings under different working conditions; converting the one-dimensional vibration signal of the motor bearing into a two-dimensional recurrence plot by adopting arecurrence plot algorithm; carrying out shearing and thumbnail preprocessing on all recursive graphs; marking fault classification labels on all recursive graphs, and dividing into a training set anda test set; constructing a multilayer convolutional neural network model; inputting the training set into the model, performing iterative learning training, and optimizing multilayer convolutional neural network parameters until the multilayer convolutional neural network model converges; deploying the trained multi-layer convolutional neural network model, inputting a test set, determining faultclassification, evaluating diagnosis accuracy, adjusting a model structure, and optimizing parameters to be optimal; storing the optimal model; and collecting one-dimensional vibration signals of themotor bearing under a new working condition, converting the signals into a two-dimensional recursive graph, preprocessing the two-dimensional recursive graph, inputting the two-dimensional recursive graph into the optimal model, and carrying out fault classification and identification on motor bearing operation.

The invention discloses an OLTC fault diagnosis method based on recurrence plot and recurrence quantitative analysis, and the method comprises the steps: 1, enabling a vibration detection probe to beattached to the top end of the box wall of an on-load tap-changer, and collecting vibration signals generated in the action process of the on-load tap-changer in different states, such as normal state, contact looseness, contact abrasion, contact burnout and the like; 2, performing phase space reconstruction on the vibration signal from the recursion characteristic of the vibration signal; 3, carrying out qualitative and quantitative analysis on the phase trajectory by respectively adopting a recursive graph (RP) and a recursive quantitative analysis (RQA); And 4, carrying out fault diagnosisaccording to qualitative and quantitative analysis results. The working state of the transformer on-load tap-changer can be monitored in real time, and the requirement for real-time fault diagnosis ofthe transformer on-load tap-changer is met. Data support and theoretical basis are provided for targeted maintenance, and manpower, material resources and time are prevented from being wasted.

There is provided a mechanism for generating random numbers following the normal distribution which does not generate a sequence correlation. The mechanism for generating random numbers following the normal distribution includes: a random recurrence plot generating mechanism 1 for generating random recurrence plots; and a recurrence plot-time series converting mechanism 4 for converting the random recurrence plots from the random recurrence plot generating unit 1 into time series, wherein Gaussian random numbers are generated by the recurrence plot-time series converting mechanism 4.

The invention provides a structural damage identification monitoring method based on a vibration signal space real-time recurrent graph convolutional neural network. The method comprises the followingsteps: S1) establishing a numerical model and generating external excitation such as wind excitation load; S2) preparing structure numerical models of different damage positions and damage degrees, and loading wind excitation load and other excitation to the numerical models; S3) generating a corresponding recurrence plot sample according to the time-history acceleration response; and S4) training and testing the convolutional neural network of the sample. The method has the advantages that the method is used for carrying out lossless damage identification on a structure in the field of civilengineering, a corresponding recursive graph generated by acceleration responses of multiple points on the structure is proposed to serve as an analysis object, and feature extraction is carried outby adopting a convolutional neural network. Compared with a traditional machine learning algorithm, the convolutional neural network has congenital advantages in feature extraction of two-dimensionaland above high-dimensional data, the training efficiency and generalization ability of the convolutional neural network in structural damage recognition can be effectively improved, and the convolutional neural network has good precision and low training cost.

The invention discloses a gas compressor vibration fault detection method based on a recurrence plot and a deep convolutional network. The method comprises the steps of firstly, constructing a compressor aerodynamic instability vibration fault database, selecting existing experimental data, and obtaining a large number of fault databases through data preprocessing; secondly, selecting a two-dimensional image conversion method; converting the one-dimensional vibration signal into a two-dimensional image, finally, selecting a deep convolutional neural network Inception V3 model, keeping a modelfeature extraction link unchanged by adopting transfer learning, and performing fault feature extraction and detection on a two-dimensional image recursive graph by adjusting a feature integration link structure. According to the method, the two-dimensional image reflects the phase space manifold in the time sequence internal dynamic system, the dynamic characteristics of the system are disclosed,and the method is suitable for carrying out characteristic analysis on the non-stationary nonlinear time sequence. Advantages in machine vision are applied to the field of non-vision, and mechanicalvibration fault detection is performed by adopting a deep convolutional neural network Inception V3 model so that huge workload of manual feature design can be reduced.

The invention belongs to the technical field of medical treatment. The invention discloses an electroencephalogram signal recognition method and system combining a recurrence plot and a CNN, and the method comprises the steps: decomposing a motor imagery electroencephalogram signal into intrinsic mode functions of different scales through the empirical mode decomposition of preprocessed electroencephalogram signal data, calculating a multi-scale recurrence plot of an intrinsic mode component of each scale, and obtaining a first-stage feature; regarding the reconstructed multi-scale recurrenceplot as image features of left and right hand EEG signals, regarding the multi-scale recurrence plot features as input of a convolutional neural network, classifying and recognizing the recurrence plot through the convolutional neural network, and extracting second-level features capable of better expressing motor imagery electroencephalogram signals from the first-level features. The electroencephalogram signal recognition rate is high, and electroencephalogram signals can be better recognized; the delay time result determined by the mutual information method is more accurate. The ReLU activation function is adopted, and when the input is a positive number, the problem of gradient saturation does not exist.

The invention discloses a transformer winding state diagnosis method based on a dynamic nonlinear characteristic sequence. The method comprises the following steps: acquiring vibration signals under different load currents by arranging a vibration sensor of an oil tank of a transformer, and acquiring current signals; extracting nonlinear ratio characteristics from a relation between the collectedcurrent and vibration by utilizing a cross recurrence plot analysis method, and computing a nonlinear characteristic sequence changed along with the load current; establishing hidden Markov models corresponding to different states, and training the HMM models by utilizing the nonlinear characteristic sequence; and for the to-be-detected transformer winding, inputting the corresponding nonlinear characteristic sequence into the trained models, comparing the probabilities obtained by different models, thereby acquiring an actual state result of the transformer winding.

The invention relates to a CFRP porosity ultrasonic characterization method based on ultrasonic backscattered signal recurrence quantification analysis, and belongs to the technical field of nondestructive detection. The method comprises the steps that an ultrasonic signal collecting system composed of an ultrasonic flaw detector, a delay block probe and a digital oscilloscope is adopted, an ultrasonic backscattered signal is collected, and phase-space reconstruction is conducted on the ultrasonic backscattered signal. Distance between every two time vectors in a phase space is calculated, a reference threshold value is set, a two-dimensional recursive matrix is obtained, and the two-dimensional recursive matrix is visualized to obtain a Recurrence Plot (RP). An RP graph is subjected to quantitative calculation through RQA quantitative index Recurrence Rate (RR), the correlation between the CFRP porosity P and RR is established finally, RR=aP-1, and characterization of the porosity is achieved. Compared with an ultrasonic attenuation method, the limitation that when a bottom echo does not exist or is weak, the porosity cannot be characterized is overcome through the method, and a good popularization prospect is achieved.

The invention discloses an intelligent monitoring method and system for loosening positioning of a steel structure bolt group. The method comprises the following steps: collecting multi-channel stress wave signals under different working conditions through a piezoelectric sensing monitoring device; preprocessing the stress wave signals; performing phase-space reconstruction on the preprocessed stress wave signals according to an improved multivariable recurrence plot algorithm to obtain a multivariable recurrence plot of a corresponding working condition; training a multi-attention mechanism improved convolutional neural network model through the multivariable recurrence plot of the corresponding working condition; and performing steel structure bolt group loosening positioning through the convolutional neural network model improved by the multi-head attention mechanism. A recursive analysis method is introduced into the field of piezoelectric active sensing, multi-sensor information fusion is realized, a multivariable recursive plot and an improved convolutional neural network are combined to realize steel structure bolt group loosening positioning, and the diagnosis precision of the model is improved.

The invention relates to a device for detecting external insulation of transmission lines and a control method thereof. The device comprises an inclined insulated support frame which supports a sample. Upper part and lower part of the sample are equipped with a high-voltage end electrode and a ground electrode respectively, wherein the high-voltage end electrode is connected with one output end of a transformer through a resistor R. A primary coil of the transformer is connected with ac power supply through a voltage regulator. A water-dripping device is arranged above the sample and an antenna is arranged in front of a sample surface, wherein the antenna is connected with a filter amplifier circuit. The filter amplifier circuit is connected to a computer through a data acquisition card. The method comprises: putting the sample on the insulated support frame and connecting the sample to the high-voltage end electrode and the ground electrode, electrifying the sample, dripping water upon the sample surface, receiving electromagnetic wave signals, drawing a three-dimension spectrum of the electromagnetic wave signals, calculating a vacancy rate lambda with a formula, drawing a two-dimension spectrum, drawing a discharging recurrence plot and calculating a recurrence rate, and determining a hydrophobicity level of the sample. The device and the method can decrease external interference and provide high accuracy, and can visually, clearly and accurately reflect changing conditions of hydrophobicity of the insulator surface.

The invention discloses a multi-threshold recurrence plot calculation method and device, and is characterized in that recursion points are inquired by a nearest neighbor point search method based on a data query tree, a multi-threshold recurrence plot is stored in a compressed format, and diagonals in different threshold recurrence plots can be recurred and searched from the multi-threshold recurrence plot. The method comprises the following specific steps: (1) establishing and initializing a recurrence plot data structure; (2) carrying out recurrence update on the data query tree; (3) searching all nearest neighbor points in a neighbourhood which takes a current point as the center and a maximum threshold value Epsilon Q as a radius in the query tree as the recurrence points, taking a formula as a reference threshold value to quantize the recurrence points, wherein a quantization distance is used for forming the matrix elements of the multi-threshold recurrence plot; (4) recording all recurrence points into the recurrence plot record; repeating the step (2) to the step (4) to search all the recurrence points; and (5) in a multi-threshold recurrence plot data record, inquiring and recording the diagonals in the different threshold recurrence plots. When only a recurrence unit in an area (TL is greater than or equal to 1 and smaller than or equal to i-j, and TH is greater than or equal to i-j and smaller than N-1) is required, the length of the query tree is not greater than NW=TH-TL+1.

The invention relates to a pulse wave signal identification and classification method based on frequency domain double-feature fusion. The method comprises the following steps: acquiring original pulse wave data of a patient, and firstly carrying out time domain pulse data preprocessing; converting the preprocessed time-domain pulse data into frequency-domain pulse data and then respectively converting the frequency-domain pulse data into a one-dimensional cepstrum coefficient and a two-dimensional recurrence plot; extracting features of the recurrence plot through a Densenet model to obtain two-dimensional image features; extracting features of the cepstrum coefficient through a CNN model to obtain one-dimensional data features; sending the one-dimensional data features and the two-dimensional image features to a feature fusion module for fusion; and sending the fusion features to a classification module for identification and classification. The high-dimensional nonlinear characteristic and the non-periodic characteristic in the pulse signal are fully considered, so that relatively deep information in the pulse characteristic can be obtained; the problem that the classification accuracy of some diseases is low due to the fact that the high-dimensional nonlinear characteristic and the non-periodic characteristic of the pulse signals are not considered in an existing pulse wave signal classification and identification method is solved.

The invention relates to a repeated-melody-based music segmentation boundary identification method, and belongs to the technical field of audio signal processing. The method comprises the following steps: 1) extracting chroma features of an audio, performing zero filling at the head and the tail, aggregating every adjacent N frames to form a new frame vector, and forming a new frame feature vectorsequence by all frame vectors; 2) calculating the Euclidean distance between each frame vector in the frame feature sequence and other frame vectors to obtain a self-similarity matrix S; 3) based onthe self-similarity matrix S, obtaining a set Ni of nearest neighbor frames of the ith frame vector, and obtaining a recurrence plot R of the self-similarity matrix S; 4) performing time delay processing on the recurrence plot R to obtain a time delay matrix L; 5) performing line segment normalization and denoising on the L, and performing reverse time delay processing to obtain a recurrence plotR', and 6) detecting all line segments and performing line segment clustering, and performing processing in sequence from the cluster with the most line segments to obtain a music segmentation boundary point set B. The identification capability for repeated melodies in music can be improved, and the music can be segmented in a shorter time.

The invention relates to a one-way valve fault diagnosis method based on total variation noise reduction and RQA, and belongs to the technical field of mechanical fault diagnosis and signal processing. The method comprises the steps of firstly, using a total variation noise reduction method for reducing noise of vibration signals, and increasing the signal-to-noise ratio; then, drawing a recurrence plot for the signal after noise reduction, extracting nonlinear characteristic parameters in the recurrence plot through a recursive quantitative analysis (RQA) method, carrying out sensitivity analysis on the extracted characteristic parameters through a compensation distance evaluation technology, and finding out characteristic parameters with higher sensitivity to form a characteristic vector; and finally, inputting the obtained feature vector into a weighted K nearest neighbor classifier (KNN) to complete the identification of the fault type of the one-way valve. Compared with the priorart, the method mainly solves the problems of low signal-to-noise ratio and unobvious fault representation caused by the fact that the one-way valve vibration signal is submerged by strong backgroundnoise, effectively improves the accuracy of one-way valve fault diagnosis, and has a certain engineering application value.

The invention discloses a driving style recognition model based on a recurrence plot and a convolutional neural network, a lane changing decision model and a decision method, and the method comprises the steps: introducing a driving style as a comprehensive judgment basis, and obtaining an accelerator pedal, a brake pedal and a steering wheel corner signal; acquiring the driving environment of a current vehicle and the driving information of surrounding vehicles through a vehicle-mounted camera and a laser radar; sampling the data to the same frequency, performing normalization processing, and generating a recurrence plot of the own vehicle and the surrounding vehicles from the data by using a recurrence plot generation algorithm; and then, inputting the recurrence plot into the convolutional neural network to obtain a driving style corresponding to each vehicle; and finally, introducing a decision-making cost function, comprehensively considering driving style factors of the vehicle and surrounding vehicles, and establishing a lane changing decision-making method by adjusting a weight coefficient of the decision-making cost function, so that a lane changing decision-making result has tendency and fits the driving style of the vehicle; and humanized, safe and reasonable driving operation is realized.

The invention provides an unbalanced load data type identification method based on VAE preprocessing and RP-2DCNN. The method comprises the following steps: step 1, acquiring load data; step 2, preprocessing the load data; step 3, carrying out balance processing on the minority class of load data by using a variational auto-encoder; 4, converting the load data into a two-dimensional recurrence plot without a threshold value by using a recurrence plot algorithm; and 5, establishing a two-dimensional convolutional neural network, training and optimizing the two-dimensional convolutional neural network to obtain a load data type identification model, and inputting the recurrence plot into the load data type identification model to obtain a classification result of the load data.

The invention discloses a pile damage identification method and device based on a convolutional neural network and a medium, and the method comprises the steps: building a plurality of models for a to-be-detected pile structure according to the condition attributes of a to-be-detected pile, solving the models, and generating a speed time history curve of a preset position point on the to-be-detected pile, and generating a speed-time history recursion plot based on the speed-time history curve, inputting the speed-time history recursion plot into the neural network model for detection, and outputting a pile damage parameter evaluation result. The pile structure of the to-be-detected pile is modeled according to the condition attributes of the to-be-detected pile, and the overall condition of the to-be-detected pile can be analyzed and known more comprehensively. And the neural network model has extremely high non-linear large-scale parameter parallel analysis and processing capability, and can better process the complex damage identification problem in the pile-soil structure and improve the accuracy of pile damage identification in combination with the convolutional neural network processing speed time-history recurrence plot.