Detailed Description
The following description of the embodiments of the present invention will be made apparent and fully in view of the accompanying drawings, in which some, but not all embodiments of the invention are shown. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The invention will now be described in further detail with reference to specific examples thereof in connection with the accompanying drawings.
Example 1
As shown in fig. 1, a multi-node edge sensing sampling system based on environmental monitoring is provided in a first embodiment of the present invention, which includes a central transmitting unit 10, a data receiving module 20, a mixing module 30, a processing and screening module 40, and an analyzing and warning module 50;
The central transmitting unit 10 is used for transmitting radio frequency signals in real time in a current noise place, the data receiving module 20 is used for receiving radio frequency signals transmitted by the central transmitting unit in real time in the current noise place, the mixing signal 30 is used for mixing the radio frequency signals with sound signals generated in the current noise place to obtain mixing signals, the processing and screening module 40 is used for carrying out Fourier transform processing and screening processing on the mixing signals to obtain noise signals, the analyzing and warning module 50 is used for carrying out analysis on the basis of the noise signals to obtain noise levels, storing and displaying the noise levels and sending warning signals to current staff on the basis of the noise levels.
The sampling system adopts a central setting method, stable radio frequency signals or induction signals are sent to surrounding space, each person wears an inductor (namely a data receiving module 20) with a signal receiver function, and the inductor is used for inducting and receiving signals sent by a central transmitting unit 10, so that the central multi-node induction sampling system can identify signals at different positions of a plurality of nodes (each node corresponds to an individual staff), real-time noise receiving measurement of the current node can be calculated, and after the noise contact quantity of the individual staff at the positions of the plurality of nodes is received on a large scale, the noise (productive noise) receiving measurement of different staff (different nodes) in the same place can be compared, thereby providing a technical basis for the subsequent objective noise (productive noise) receiving measurement of evaluating the nodes all the year round or a certain time period.
Because a plurality of central emission units 10 are arranged in the place, the central emission units 10 are connected nearby by only needing personnel to walk to different position points, and the central emission units 10 are used for completing noise level acquisition, so that each person does not need to bear a measuring sampler with huge volume.
Preferably, the processing and screening module 40 includes a transformation processing module 41, a feature extraction and screening module 42, a matching module 43, and a screening analysis module 44;
The transformation processing module 41 is configured to perform fourier transformation processing based on the mixed signal to obtain spectrum information Sy (f);
The feature extraction and screening module 42 is configured to perform a time domain feature extraction operation on the mixed signal in the observation period T to obtain time domain feature information;
The matching module 43 is configured to obtain a plurality of frequency spectrum ranges based on the target frequency spectrum information Sy (f'), and obtain a typical mechanical noise frequency spectrum Scoremech and a typical mechanical noise frequency spectrum Scoreclass based on matching a preset mechanical noise frequency spectrum Smech (f) and a mechanical noise-like frequency spectrum Sclass (f) by combining the frequency spectrum ranges;
The screening analysis module 44 is configured to perform screening analysis based on the typical mechanical noise spectrum Scoremech, the typical mechanical noise spectrum Scoreclass, and the time domain feature information to obtain a noise signal;
Preferably, the time domain characteristic information includes a root mean square RMS value RMS (y (t)) and an autocorrelation function Ry (τ).
Preferably, the screening analysis module 44 is specifically configured to determine whether the value of the typical mechanical noise spectrum Scoremech is greater than the typical mechanical noise spectrum Scoreclass, if so, determine that the target spectrum information Sy (f') corresponding to the spectrum range is the mechanical noise signal to be determined;
Preferably, the feature extraction and screening module 42 includes a primary screening module 421, a secondary screening module 422, and a merging module 423;
The primary screening module 421 is configured to perform primary screening on the mixed signal based on the time domain feature information to obtain first target spectrum information Sy(f')1 and a mixed signal to be screened;
The rescreening module 422 is configured to perform a rescreening operation based on the transit time dimension of the to-be-screened mixed signal and the frequency component of the to-be-screened mixed signal, to obtain second target spectrum information Sy(f')2;
The merging module 423 is configured to obtain target spectrum information Sy (f') based on the first target spectrum information Sy(f')1 and the second target spectrum information Sy(f')2.
Preferably, the rescreening module 422 includes a time window dividing module 4221, a frequency component obtaining module 4222, a first calculating module 4223, a second calculating module 4224, and a judging module 4225;
the time window dividing module 4221 is configured to divide the to-be-screened mixed signal according to a time sequence corresponding to the to-be-screened mixed signal to obtain a plurality of time windows i= { i1, i2, i3, &. IN }, where iN is an nth time window;
The frequency component obtaining module 4222 is configured to obtain frequency components j= { j1, j2, j3, &. JK } corresponding to each time window i through short-time fourier transform processing on each time window i, where jK is a frequency component of a kth time window;
the first calculating module 4223 is configured to calculate a time correlation index TC (i) based on the time-frequency matrix M;
The second calculation module 4224 is configured to perform iterative calculation based on the time-frequency matrix M and combining a time scale and a frequency scale of the time-frequency matrix M to obtain a spectrum stability index SF;
The determining module 4225 is configured to determine whether the comprehensive scoring function Score (i) is greater than or equal to a preset comprehensive scoring function threshold, and if yes, determine that the mixing signal to be filtered corresponding to the time window is second target spectrum information Sy(f')2.
Preferably, the second computing module 4224 includes an initializing module 42241, a matrix constructing module 42242, a third computing module 42243, a fourth computing module 42244, a fifth computing module 42245, a sixth computing module 42246, a seventh computing module 42247, and a judging output module 42248;
The initialization module 42241 is configured to construct an initial spectrum stability matrix SSM (0) based on the time-frequency matrix M, calculate a density index D (i) based on the time-frequency matrix M, and initialize iteration parameters;
the matrix construction module 42242 is configured to, for the iteration parameters, include a smoothing coefficient α, an initial spectrum stability index SF (i), a time scale parameter τt, a frequency scale parameter τf, a weight coefficient β1, a weight coefficient β2, a weight coefficient β3, a learning rate γ, an iteration counter, and an iteration maximum number threshold, where the iteration number of the iteration counter is initially 0;
The third calculation module 42243 is configured to construct a time domain similarity matrix U based on the time window and the time scale parameter τt, and construct a frequency domain similarity matrix V based on the frequency component and the frequency scale parameter τf;
the fourth calculation module 42244 is configured to iteratively update the spectrum stability matrix, i.e., add 1 to the iteration number of the iteration counter to obtain a current iteration number, and calculate an updated spectrum stability matrix SSM (z+1) based on the smoothing coefficient α, the initial spectrum stability matrix SSM (0), the time domain similarity matrix U, and the frequency domain similarity matrix V;
The fifth calculation module 42245 is configured to calculate, based on the updated spectrum stability matrix SSM (z+1), by solving a eigenvalue equation, to obtain a target eigenvector C (i);
the sixth calculation module 42246 is configured to calculate an updated spectrum stability indicator SFnew (i) based on the target feature vector C (i), the density indicator D (i), and the initial spectrum stability indicator SF (i);
the seventh calculation module 42247 is configured to calculate a target convergence threshold value under the current iteration number based on the current iteration number in combination with a preset initial convergence threshold value, and calculate a comprehensive convergence value TC based on the updated spectrum stability index SFnew (i) and the updated spectrum stability matrix SSM (z+1);
The judging and outputting module 42248 is configured to judge whether the integrated convergence value is less than or equal to the target convergence threshold, if yes, output the updated spectrum stability index SFnew (i) as the target spectrum stability index SF, if not, judge whether the current iteration number is greater than or equal to a preset maximum iteration number threshold, if yes, output the updated spectrum stability index SFnew (i) as the target spectrum stability index SF, if not, update the smoothing coefficient α based on the updated spectrum stability matrix SSM (z+1) to obtain an updated smoothing coefficient α ', take the updated smoothing coefficient α' as the smoothing coefficient α, return to the iterative updating spectrum stability matrix operation, and re-execute until the target spectrum stability index SF is obtained.
Preferably, the seventh calculating module 42247 is specifically configured to calculate a relative rate of change RC of the spectrum stability matrix based on the updated spectrum stability matrix SSM (z+1);
And calculating based on the relative change rate RC of the frequency spectrum stability matrix and the relative change rate SC of the stability index to obtain a comprehensive convergence value TC.
Preferably, the judging output module 42248 is specifically configured to calculate the updated smoothing coefficient α' based on the relative change rate RC of the spectrum stability matrix.
Preferably, the updated smoothing coefficient α' is calculated by:
α'=α×(1+γ×RC);
wherein, gamma is the learning rate.
In summary, the multi-node edge induction sampling system based on environmental monitoring provided by the embodiment of the application firstly transmits radio frequency signals through the central transmitting unit 10 to serve as a stable signal source to provide a reference standard for subsequent noise analysis, then the data receiving module 20 receives the radio frequency signals from the central transmitting unit, and then the mixing module 30 generates mixing signals by mixing the received radio frequency signals with sound signals in noise places to lay a foundation for subsequent noise extraction and analysis;
In the processing and screening stage, a transformation processing module 41 in a processing and screening module 40 performs Fourier transformation on the mixed signal to obtain frequency spectrum information, a characteristic extraction and screening module 42 performs time domain analysis on the mixed signal based on time domain characteristic extraction, extracts time domain characteristic information and screens to obtain target frequency spectrum information, a matching module 43 obtains a plurality of frequency spectrum ranges according to the target frequency spectrum information and matches with a preset mechanical noise frequency spectrum and a similar mechanical noise frequency spectrum to determine typical mechanical noise and similar mechanical noise frequency spectrum, and a screening analysis module 44 performs further analysis by utilizing the frequency spectrum information and the time domain characteristic information to accurately identify noise signals;
in the rescreening process, the rescreening module 422 firstly divides the mixed signal into a plurality of time windows according to the time sequence through the time window dividing module 4221, and performs short-time fourier transform through the frequency component obtaining module 4222 to obtain the frequency component of each time window; then, the first calculation module 4223 calculates a time-related index based on the time-frequency matrix M, and the second calculation module 4224 performs iterative calculation by combining the time and frequency scales of the time-frequency matrix to obtain a spectrum stability index SF, and calculates a comprehensive scoring function based on the indexes;
Further, the second calculation module 4224 constructs an initial spectrum stability matrix SSM (0) through the initialization module 42241 and calculates a density index, the matrix construction module 42242 sets iteration parameters including a smoothing coefficient, a time scale, a frequency scale, and the like, and starts iterative calculation, in the iterative process, the spectrum stability matrix is continuously updated, and the time domain similarity matrix U and the frequency domain similarity matrix V are respectively updated through the third calculation module 42243 and the fourth calculation module 42244;
Finally, judging whether the preset convergence condition is met or not by the judging output module 42248 according to the comprehensive convergence value, and outputting a final spectrum stability index SF if the preset convergence condition is met, if the preset convergence condition is not met, continuing iteration, and updating the smoothing coefficient until a final result is obtained;
Through the layering operation, the system can accurately extract relevant components of the production noise from the complex mixed signals, evaluate the noise level and timely send out warning signals, so that the health and safety of staff are effectively ensured.
Example two
As shown in fig. 2, a second embodiment of the present invention provides a multi-node edge sensing sampling method based on environmental monitoring, which includes the following operation steps:
s10, when workers are in the current noise place, the portable sensors worn by the workers receive radio frequency signals sent by the central transmitting unit in real time;
In the embodiment of the application, the central transmitting unit is arranged in the noise place to transmit the stable radio frequency signal which covers the whole noise place in real time, when the staff is in the noise place, the portable sensor worn by the staff is positioned in the radio frequency signal coverage of the central transmitter, and the portable sensor can receive the radio frequency signal in the current coverage at any time and any place, and the radio frequency signal is a stable signal source and provides a reference signal for subsequent processing analysis.
Step S20, mixing the radio frequency signal with a sound signal generated in the current noise place to obtain a mixed signal;
It should be noted that in the above embodiment of the present application, the sound signal may not be directly processed by computer operation, but the portable sensor in the embodiment of the present application mixes the radio frequency signal and the sound signal through the nonlinear element to generate a new frequency component, that is, the above mixing signal;
it should be noted that, the sound signal in the above embodiment of the present application includes a noise signal that is generated by mechanical impact, friction, rotation, etc. in the current noise location and has a significant impact on human health, and the sound signal also includes a similar mechanical noise that is generated by a worker during working and walking and is different from the noise signal;
step S30, carrying out Fourier transform processing and screening processing on the mixed signals to obtain noise signals;
and step S40, analyzing the noise signal to obtain a noise level, storing and displaying the noise level, and sending out a warning signal to the current staff based on the noise level.
It should be noted that, in the embodiment of the application, through the cooperation of the portable sensor worn by the staff and the central transmitting unit, the noise condition in the productive noise place is monitored in real time, the noise signal is analyzed through the frequency mixing technology, and finally the noise level information is provided and the warning is sent out;
In the specific operation process, the radio frequency signal transmitted by the central transmitting unit is used as a stable signal source, a stable and consistent signal source is provided for the subsequent mixing process, the system is ensured to accurately extract and analyze the noise signal, the system can extract and process noise signals with different frequencies by mixing the noise signal acquired by the portable sensor at any time and any place with the reference radio frequency signal, a data basis is provided for the subsequent noise analysis, further, components related to productive noise are extracted from the mixed signal, the system is ensured to accurately identify the noise signal influencing human health from the complex mixed signal, a foundation is laid for the subsequent noise level evaluation, the noise level is obtained through analysis according to the intensity or frequency spectrum characteristics of the noise signal, when the noise level exceeds a set threshold value, the system timely sends out an alarm signal to remind a worker to pay attention to noise hazard, the worker is prevented from being exposed in a harmful noise environment for a long time, the system can monitor the noise environment, and can also safely warn the worker to work within a healthy range by providing real-time noise level evaluation and warning.
Specifically, in step S30, fourier transform processing and filtering processing are performed on the mixed signal to obtain a noise signal, which includes the following steps:
Step S31, carrying out Fourier transform processing based on the mixed signal to obtain spectrum information Sy (f);
It should be noted that, in the above embodiment of the present application, in the spectrum information Sy (f) obtained by fourier transform processing of the mixed signal, the frequency component of the noise may be extracted, and the amplitude distribution of different frequency regions may be analyzed; because the mechanical noise and the similar mechanical noise (similar mechanical noise is noise generated by friction, conversation and the like of clothes or appliances in the working and walking processes of workers and near a portable sensor) are different in frequency spectrum characteristics, the mechanical noise is usually represented as a relatively stable frequency component in frequency spectrum, and the similar mechanical noise is usually relatively strong in instantaneous variability in frequency spectrum, relatively disordered in frequency distribution and periodic;
Step S32, performing time domain feature extraction operation on the mixed signal in an observation period (namely, a continuous time period) to obtain time domain feature information, and screening the mixed signal based on the time domain feature information to obtain target frequency spectrum information Sy (f');
The time domain feature information includes a root mean square RMS value RMS (y (t)) and an autocorrelation function Ry (τ);
the root mean square RMS value RMS (y (t)) is calculated by:
in the formula,T is the observation period (i.e., the continuous time period), otherwise known as the time length of the mixed signal measurement);
The autocorrelation function Ry (τ) is calculated in the following manner:
in the formula,Τ is a time delay (or time offset, which represents a time difference between two time points of the mixed signal), and y (t) is the mixed signal at time point t;
It should be noted that, in the above embodiment of the present application, the power or energy of the signal is measured by calculating the RMS value RMS (y (t)), which represents the average level of the signal amplitude, and in general, the mechanical noise usually presents a relatively stable RMS value RMS (y (t)), while the similar mechanical noise usually has higher instantaneous volatility, so that when the RMS value RMS (y (t)) exceeds the preset RMS value threshold, the signal of a certain segment (i.e. the above time period) in the mixed signal is proved to be similar to the mechanical noise, and when the RMS value RMS (y (t)) does not exceed the preset RMS value threshold, the signal of a certain segment (i.e. the above time period) in the mixed signal is proved to be the mechanical noise, i.e. the above target spectrum information Sy (f');
The correlation of the mixed signal y (t) under different time delays τ is described by the calculation of the autocorrelation function Ry (τ), so that the periodicity of the signal is determined, the mechanical noise usually has a relatively stable autocorrelation characteristic, and the similar mechanical noise usually shows an irregular or relatively transient autocorrelation characteristic, so that when the autocorrelation function Ry (τ) exceeds a preset autocorrelation function, the signal of a certain segment (i.e., the time period) of the mixed signal is proved to be similar to the mechanical noise, and when the autocorrelation function Ry (τ) does not exceed the preset autocorrelation function, the signal of a certain segment (i.e., the time period) of the mixed signal is proved to be the mechanical noise, i.e., the target spectrum information Sy (f').
Step S33, acquiring a plurality of frequency spectrum ranges based on the target frequency spectrum information Sy (f'), and matching based on a preset mechanical noise frequency spectrum Smech (f) and a similar mechanical noise frequency spectrum Sclass (f) by combining the frequency spectrum ranges to obtain a typical mechanical noise frequency spectrum Scoremech and a typical similar mechanical noise frequency spectrum Scoreclass;
The typical mechanical noise spectrum Scoremech is calculated by:
The typical mechanical noise spectrum Scoreclass is calculated by the following steps:
wherein, fmax and fmin are frequency spectrum ranges, df is frequency interval;
It should be noted that, in the embodiment of the present application, each peak top value fmax and peak valley value fmin of the target spectrum information Sy (f') are first obtained, then the peak top value fmax and peak valley value fmin are ordered according to a time sequence, and then a spectrum range is constructed by adjacent peak top value fmax and peak valley value fmin;
Step S34, screening and analyzing based on the typical mechanical noise spectrum Scoremech, the typical mechanical noise spectrum Scoreclass and the time domain characteristic information to obtain a noise signal;
Specifically, in the above embodiment of the present application, the values of the typical mechanical noise spectrum Scoremech and the typical mechanical noise spectrum Scoreclass are further selected by determining whether the value of the typical mechanical noise spectrum Scoremech is greater than the typical mechanical noise spectrum Scoreclass, if yes, determining the target spectrum information Sy (f ') corresponding to the spectrum range as the mechanical noise signal to be determined, if not, determining whether the value of the typical mechanical noise spectrum Scoremech is less than the typical mechanical noise spectrum Scoreclass, if yes, determining the target spectrum information Sy (f') corresponding to the spectrum range as the mechanical noise signal to be determined;
It should be noted that, in the embodiment of the application, through twice screening, firstly, signals are converted from time domain to frequency domain through fourier transformation, a foundation is laid for subsequent noise analysis, the expression forms of mechanical noise and similar mechanical noise in the frequency domain are different, the fourier transformation can help a system to quickly identify the differences, the frequency composition of the signals can be directly observed through frequency domain information, the complexity of time domain analysis is simplified, the frequency characteristics of the noise can be more intuitively identified, and further, the mechanical noise (energy stability) and the similar mechanical noise (energy fluctuation is large) can be effectively distinguished according to the energy level of the signal reflected according to root mean square RMS value, and the characteristics are helpful for quickly screening the noise in time. The system can further confirm the periodic characteristics of the signals through autocorrelation analysis, enhance the accuracy of noise classification, further define the frequency spectrum ranges of the mechanical noise and the mechanical noise-like by analyzing the frequency spectrum peak value and the peak valley value of the typical noise in the execution process of the step S33-the step S34, and define the noise types more accurately through peak sorting and construction of the frequency spectrum ranges, thereby avoiding rough classification of the frequency ranges, improving the accuracy and reliability of noise identification, and finally confirm the noise types through comparison of the frequency spectrum of the typical mechanical noise and the frequency spectrum of the mechanical noise-like, such multi-level screening, not only considers the frequency spectrum characteristics, but also synthesizes the time domain analysis results, so that the noise classification is more accurate, and furthest reduces the classification errors, so that the final noise identification result is more reliable.
In the implementation process of the embodiment of the present application, the skilled person also finds that in the machine production process, the transient noise is generated due to the short-term friction of the machine caused by starting the test machine, and the transient noise is included in the category of the mechanical noise when the target spectrum information Sy (f ') is obtained through the time domain feature information, so that the target spectrum information Sy (f ') is obtained through the time domain feature information, and further analysis and recognition are required to obtain the mechanical noise, so as to obtain more accurate target spectrum information Sy (f ').
Specifically, in step S32, the filtering of the mixed signal based on the time domain feature information to obtain the target spectrum information Sy (f'), includes the following steps:
step S321, performing primary screening on the mixed signal based on the time domain characteristic information to obtain first target frequency spectrum information Sy(f')1 and a mixed signal to be screened;
It should be noted that, in the above embodiment of the present application, the RMS (y (t)) is smaller than or equal to the preset RMS threshold, and the mixing signal corresponding to the time period when the autocorrelation function Ry (τ) is smaller than or equal to the preset autocorrelation function is determined as the first target spectrum information Sy(f')1, and the mixing signal except the first target spectrum information Sy(f')1 is determined as the mixing signal to be screened so as to perform the screening operation again, and the above primary screening is the first screening operation and also the coarse screening operation, so as to screen the first target spectrum information Sy(f')1 which can be more directly identified as the target spectrum information Sy (f'), and screen to obtain the mixing signal to be screened, and provide a data basis for the subsequent second screening (or fine screening) operation.
Step S322, performing rescreening operation based on the passing time dimension of the mixed signal to be screened and the frequency component of the mixed signal to be screened to obtain second target frequency spectrum information Sy(f')2;
It should be noted that, in general, the mechanical noise (i.e. noise generated by friction, talking, etc. of clothes or devices during work and walking of a worker and near the portable sensor) will be represented as continuous, low-frequency (representing low in the audio frequency band and the vibration frequency band, and the unit is Hz) mixing signal, and the test machine is started, so that the transient noise generated by the mechanical transient friction is a high-frequency mixing signal, in addition, the mechanical noise is usually represented as continuous frequency in the appearance frequency and higher than the transient noise generated by the mechanical transient friction.
Step S323 obtains target spectrum information Sy (f') based on the first target spectrum information Sy(f')1 and the second target spectrum information Sy(f')2.
The embodiment of the application firstly utilizes the time domain characteristics (namely root mean square RMS value and autocorrelation function) to carry out primary screening on the mixed signals, determines to obtain partial mixed signals which can be directly identified as target frequency spectrum information (namely first target frequency spectrum information), determines the rest mixed signals as signals to be screened, rapidly filters out signals conforming to the target characteristics through simple time domain characteristic analysis, and provides a basis for subsequent screening;
However, because the first target spectrum information can be obtained by direct screening, but the second target spectrum information Sy(f')2 is important reference information, which contains a part of short-term mechanical noise which is not identified into the target spectrum information and is needed to be further shared and screened, in the embodiment of the application, the mixed signal to be screened is further screened again, particularly by analyzing the time dimension and the frequency component so as to more accurately distinguish the mechanical noise from the similar mechanical noise, thereby extracting the second target spectrum information, and the high-frequency short-term mechanical noise is accurately identified by analyzing the thinned time dimension and the frequency component, so that the screening precision is further improved, and finally, the first target spectrum information and the second target spectrum information are combined to obtain the final target spectrum information, and the result of the two screening is finally combined to form the accurate target spectrum information, thereby ensuring the comprehensiveness and the accuracy of the final result.
Specifically, in step S322, a rescreening operation is performed based on the transit time dimension of the mixed signal to be screened and the frequency component of the mixed signal to be screened, so as to obtain second target spectrum information Sy(f')2, which includes the following steps:
step S3221, dividing the mixed signal to be screened according to a time sequence corresponding to the mixed signal to be screened to obtain a plurality of time windows i= { i1, i2, i3,. IN }, wherein iN is an Nth time window;
Step S3222, obtaining frequency components j= { j1, j2, j3, &. JK } corresponding to each time window i through short-time Fourier transform processing on each time window i, wherein jK is the frequency component of the Kth time window;
step S3223, constructing a time-frequency matrix M based on the time window i and the frequency component corresponding to the time window i;
where miN,jK represents the amplitude of the kth frequency component j corresponding to the nth time window i;
It should be noted that, in the embodiment of the present application, the time domain and frequency domain characteristics of the signal can be captured simultaneously in a matrix form, so as to analyze the distribution characteristics of noise on a time-frequency plane.
Step S3224, calculating based on the time-frequency matrix M to obtain a time correlation index TC (i);
the calculation mode of the time correlation index TC (i) is as follows:
TC(i)=corr(M[i,:],M[i+1,:]);
Wherein corr () is a correlation coefficient calculation function, M [ i ]: is an ith column element in the time-frequency matrix;
It should be noted that, in the embodiment of the application, the correlation of the frequency spectrum of the mixing signal to be screened in the neighborhood time window is measured by calculating the time correlation index, the frequency spectrum characteristics of the mechanical noise have stronger correlation in the adjacent time window, and the similar mechanical noise with abrupt frequency spectrum characteristics can be identified by calculating the time correlation index.
Step S3225, performing iterative computation based on the time-frequency matrix M and combining the time scale and the frequency scale of the time-frequency matrix M to obtain a spectrum stability index SF, and performing computation of a comprehensive scoring function Score (i) based on the spectrum stability index SF and the time-related index TC (i);
the calculation mode of the comprehensive scoring function Score (i) is as follows:
Score(i)=w1*TC(i)+w2*SF(i);
Wherein w1 and w2 are weight coefficients;
the spectrum distribution of the mechanical noise should be relatively concentrated, and by calculating the spectrum stability index SF, the mechanical noise having a discrete spectrum distribution can be distinguished.
Step S3226 is to determine whether the comprehensive scoring function Score (i) is greater than or equal to a preset comprehensive scoring function threshold, if yes, determining that the mixing signal to be screened corresponding to the time window is similar to mechanical noise, screening out and not performing the next processing, and determining that the mixing signal to be screened corresponding to the time window is the second target spectrum information Sy(f')2.
By adopting the technical scheme provided by the embodiment of the application, the continuous mechanical noise and the transient noise can be effectively distinguished, the accuracy of the target frequency spectrum information is improved, the misjudgment rate is reduced, and particularly, the misjudgment of the friction noise generated at the moment of starting is performed;
According to the embodiment of the application, firstly, the mixed signal is divided into a plurality of time windows to carry out local analysis, the adverse effect caused by time-varying property possibly existing in the whole signal processing process is avoided, the signal characteristic in each time window can be independently processed to facilitate capturing of local noise mode and spectrum characteristic, according to the embodiment of the application, the change of the signal in each time period can be effectively captured, especially for non-stationary signals (such as mechanical noise), the divided signal can reflect the change in time, further, the signal of each time window is converted from time domain to frequency domain through short-time Fourier transform, so that the spectrum characteristic in each time period can be clearly displayed, further, the spectrum information of each time window can be organized together in a matrix form to facilitate carrying out comprehensive analysis on the time-frequency characteristic of the signal, the mechanical noise with abrupt change of the spectrum characteristic can be effectively distinguished, the target spectrum information can be further obtained through calculating stability and time-related index, the mechanical noise can be accurately distinguished from the mechanical noise with stable distribution, and the mechanical noise can be further distinguished, and the spectrum characteristic can be further screened according to the mechanical noise with discrete and stable distribution.
Specifically, in step S3225, the spectrum stability indicator SF is obtained by performing iterative computation based on the time-frequency matrix M and combining the time scale and the frequency scale of the time-frequency matrix M, including the following operation steps:
Step S33251, constructing an initial spectrum stability matrix SSM (0) based on the time-frequency matrix M, simultaneously calculating a density index D (i) based on the time-frequency matrix M, and initializing iteration parameters;
The iteration parameters comprise a smoothing coefficient alpha, an initial frequency spectrum stability index SF (i), a time scale parameter tau t, a frequency scale parameter tau f, a weight coefficient beta 1, a weight coefficient beta 2, a weight coefficient beta 3, a learning rate gamma (the speed of controlling parameter self-adaptive adjustment), an iteration counter and an iteration maximum frequency threshold, wherein the iteration frequency of the iteration counter is initially 0;
the initial spectral stability matrix SSM (0) is expressed as:
Where m [ iN,jK ] represents the nth time window i, the amplitude of the kth frequency component j, μK represents the mean value of the kth frequency component (represents the mean value of the frequency component over all time windows), σK represents the standard deviation of the kth frequency component (reflecting the degree of fluctuation of the kth frequency component), ε is a smoothing factor (preventing denominator from being 0);
the density index D (i) is calculated by the following steps:
D(i)=sum(exp(-||M[i,:]-M[j,:]||/σd));
Wherein M [ i ] represents the ith column element (i.e. all frequency components under the same time window i) in the time-frequency matrix M, M [ j ] represents the jth column element (i.e. different time windows corresponding to the same frequency component j) in the time-frequency matrix M, σd is the density calculation bandwidth parameter (controlling the smoothness degree of density estimation);
Step S33252 is to construct a time domain similarity matrix U based on the time window and the time scale parameter τt, and construct a frequency domain similarity matrix V based on the frequency component and the frequency scale parameter τf;
the time domain similarity matrix U is expressed as:
the frequency domain similarity matrix V is expressed as:
It should be noted that, in the above embodiment of the present application, the time domain similarity matrix U represents the degree of correlation between different time windows, and the smaller the ix-iN, the higher the time domain similarity;
Step S33253, iteratively updating a spectrum stability matrix, namely adding 1 to the iteration number of the iteration counter to obtain the current iteration number, and calculating based on the smooth coefficient alpha, the initial spectrum stability matrix SSM (0), a time domain similarity matrix U and the frequency domain similarity matrix V to obtain an updated spectrum stability matrix SSM (z+1);
the updated spectrum stability matrix SSM (z+1) is expressed as:
SSM(z+1)=α*SSM(z)+(1-α)*U*SSM(z)*V;
Wherein z is the iteration number of the last iteration, z+1 is the current iteration number, SSM (z+1) represents an updated spectrum stability matrix obtained by updating a spectrum stability matrix generated by the last iteration number under the current iteration number, SSM (z) represents an updated spectrum stability matrix under the last iteration number (SSM (0) represents an initial spectrum stability matrix when the first iteration is performed, namely when the current iteration number is 1);
Step S33254, calculating to obtain a target feature vector C (i) by solving a feature value equation based on the updated spectrum stability matrix SSM (z+1);
It should be noted that, in the embodiment of the present application, the eigenvalue equation is solved by using the updated spectrum stability matrix SSM (z+1), and then the eigenvector corresponding to the largest eigenvalue is taken as the target eigenvector C (i);
Step S33255, calculating an updated spectrum stability index SFnew (i) based on the target feature vector C (i), the density index D (i) and the initial spectrum stability index SF (i);
the calculation mode of the updated spectrum stability index SFnew (i) is as follows:
SFnew(i)=β1*SF(i)+β2*C(i)+β3*D(i);
step S33256 is to calculate the target convergence threshold under the current iteration number based on the current iteration number and the preset initial convergence threshold, and calculate the comprehensive convergence value TC based on the updated spectrum stability index SFnew (i) and the updated spectrum stability matrix SSM (z+1);
Step 33257, judging whether the comprehensive convergence value is smaller than or equal to the target convergence threshold, if yes, outputting the updated spectrum stability index SFnew (i) as a target spectrum stability index SF, if not, judging whether the current iteration number is larger than or equal to a preset maximum iteration number threshold, if yes, outputting the updated spectrum stability index SFnew (i) as a target spectrum stability index SF, if not, updating the smooth coefficient alpha based on the updated spectrum stability matrix SSM (z+1) to obtain an updated smooth coefficient alpha ', taking the updated smooth coefficient alpha' as the smooth coefficient alpha, returning to the iteration updating spectrum stability matrix operation of the step 33253, and re-executing until the target spectrum stability index SF is obtained.
It should be specifically noted that, in the embodiment of the application, an initial spectrum stability matrix SSM (0) is firstly constructed based on a time-frequency matrix M, a density index is calculated, related iteration parameters are initialized, a foundation is laid for subsequent iteration update, the spectrum stability matrix is ensured to be effectively updated in the iteration process, further, the distribution characteristics of signals in time and frequency dimensions are analyzed through a time-frequency domain similarity matrix, support is provided for subsequent spectrum stability update and feature extraction, the information of the time-frequency domain similarity matrix and the time-frequency domain similarity matrix is subjected to iteration optimization, the spectrum stability matrix gradually reflects the real stability characteristics of the signals, the time-frequency similarity matrix is utilized, the spectrum stability matrix is optimized by combining the smooth coefficient, the effective convergence of the matrix in the time and frequency dimensions is ensured, a target feature vector is extracted through a feature value equation, the spectrum stability characteristics of the signals are extracted into a concise and effective vector representation, the basis is provided for subsequent stability index update, the stability of the signals is further quantized through calculation of the updated spectrum stability index, so that the final signal characteristics conforming to the target stability are screened out, the final signal characteristics are further calculated, the convergence value and the target convergence value is further calculated, the target convergence value is effectively converged, the target convergence value is calculated, and the necessary iteration stability is avoided, and the final iteration stability is avoided.
Specifically, in step S33256, a comprehensive convergence value TC is calculated based on the updated spectrum stability index SFnew (i) and the updated spectrum stability matrix SSM (z+1), including the following steps:
Step S332561, calculating a relative change rate RC of the spectrum stability matrix based on the updated spectrum stability matrix SSM (z+1), and calculating a relative change rate SC of the stability index based on the updated spectrum stability index SFnew (i);
The calculation mode of the relative change rate RC of the spectrum stability matrix is as follows:
Wherein SSM (z) is an updated spectrum stability matrix under the previous iteration times;
the calculation mode of the stability index relative change rate SC is as follows:
Where SFnew (i) 'is the updated spectrum stability index at the last iteration (at the first iteration, SFnew (i)' is the initial spectrum stability index SF (0));
Step S332562, calculating based on the relative change rate RC of the frequency spectrum stability matrix and the relative change rate SC of the stability index to obtain a comprehensive convergence value TC;
The calculation mode of the comprehensive convergence value TC is as follows:
TC=w3×RC+w4×SC;
wherein w3 and w4 are weight coefficients, and w1+w2=1;
It should be noted that, in the embodiment of the application, the relative change rate RC of the spectrum stability matrix and the relative change rate SC of the spectrum stability index are calculated, so that the change degree of the spectrum stability matrix and the stability index in the iterative process of the system is effectively measured, the termination or continuation of the iteration is determined by integrating the convergence value TC, the calculation can be stopped in time when the set convergence standard is reached, unnecessary calculation waste is avoided, the RC and the SC provide different visual angles of signals in the aspects of local stability and global stability, the integrated convergence value TC combines the two indexes in a weighted manner, a comprehensive convergence judgment is provided, the method can be flexibly adjusted according to different signal characteristics, and the convergence progress of different stages in the iterative process can be effectively controlled.
Specifically, in step S33257, the smoothing coefficient α is updated based on the updated spectrum stability matrix SSM (z+1), so as to obtain an updated smoothing coefficient α', which includes the following operation steps:
step S332571, calculating an updated smoothing coefficient alpha' based on the relative change rate RC of the spectrum stability matrix;
The updated smoothing coefficient alpha' is calculated in the following way:
α'=α×(1+γ×RC);
Wherein, gamma is the learning rate;
It should be specifically noted that, in the above embodiment of the present application, the relative change rate RC of the spectrum stability matrix of the updated spectrum stability matrix SSM (z+1) is considered, and by combining RC and the learning rate γ, a dynamic adjustment mechanism is introduced for the smoothing coefficient, and the change degree of the spectrum stability matrix directly affects the updating of the smoothing coefficient, so that the smoothing process can be automatically adjusted according to the stability of the signal.
In summary, the multi-node edge induction sampling system and the multi-node edge induction sampling method based on the environment monitoring provided by the embodiment of the invention provide a stable and consistent signal source for the subsequent mixing process by firstly utilizing the radio frequency signal transmitted by the central transmitting unit as a stable signal source, ensure that the system can accurately extract and analyze noise signals;
Further, the components related to productive noise are extracted from the mixed signals, so that the system can accurately identify noise signals influencing human health from complex mixed signals, lay a foundation for subsequent evaluation of noise levels, analyze the noise levels according to the intensity or frequency spectrum characteristics of the noise signals, timely send out warning signals when the noise levels exceed a set threshold value, remind workers of paying attention to noise hazards and avoid long-time exposure to harmful noise environments;
further, when extracting components related to productive noise from the mixed signal, the multi-level screening considers the spectrum characteristics and synthesizes the time domain analysis result, so that the noise classification is more accurate;
Further, analysis and screening are carried out through time dimension and frequency components, so that accurate target frequency spectrum information is formed, and comprehensiveness and accuracy of a final result are ensured;
Further, the mechanical noise which is concentrated and stable in frequency spectrum distribution and the mechanical noise which is discrete in frequency spectrum distribution and is rapid in change are accurately distinguished by constructing a time-frequency matrix and calculating time correlation indexes and frequency spectrum stability;
Further, in a specific operation process, an initial spectrum stability matrix SSM (0) constructed based on a time-frequency matrix M, a calculated density index and an iteration parameter are combined with a comprehensive convergence value and a smoothing coefficient which dynamically change in the iteration process to dynamically judge and output to obtain a target spectrum stability index SF.
Finally, it should be noted that the above-mentioned embodiments are merely for illustrating the technical solutions of the present invention, and not for limiting the same, and that one skilled in the art may modify the technical solutions described in the above-mentioned embodiments or make equivalent substitutions for some or all of the technical features thereof, and these modifications or substitutions do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention.