CN107271184A - The kernel regression decomposition method and system of a kind of rolling bearing fault diagnosis - Google Patents

Abstract

The invention discloses the kernel regression decomposition method and system of a kind of rolling bearing fault diagnosis.The present invention includes motor, V belt translation axle, shaft coupling, test bearing, acceleration transducer, multi-channel data acquisition analyzer, computer；The faulty bearings are connected by motor, V belt translation axle and shaft coupling, acceleration transducer is fixed on test bearing bearing block, acceleration transducer output end is connected with multi-channel data acquisition analyzer, data after analyzer extraction after preservation by being sent to computer, and signal data is analyzed and processed on computers with reference to kernel regression decomposition method, bearing combined failure is identified, the accurate detection to rolling bearing running status is realized.The present invention can effectively gather the fault message of rolling bearing, and good with transmission performance, accuracy rate is done, and speed is fast, simple operation and other advantages, with engineering application value.

Description

Kernel regression decomposition method and system for rolling bearing fault diagnosis

technical field

The invention relates to the field of fault diagnosis of mechanical equipment, in particular to a kernel regression decomposition method and system for fault diagnosis of rolling bearings.

Background technique

Rolling bearings are important mechanical components widely used in various machines, and are also one of the most easily damaged components in machines. Rotating machinery equipment is entirely dependent on the health of rolling element bearings, accounting for almost 40-50% of equipment failures. Bearing failures can be extremely serious, and may lead to the shutdown of the entire production line, and even cause casualties. At this stage, the fault diagnosis of bearings mainly judges the operating status of bearings through manual experience or instruments. However, in the actual acquisition of vibration signals, it is often difficult to obtain vibration signals due to complex working conditions, high environmental noise, and long-term equipment working conditions. A signal with obvious fault characteristics. Therefore, signal denoising has become a key step in bearing fault signal processing. The current fault diagnosis system has limitations such as poor stability and cumbersome operation, and cannot be widely used and promoted. In order to solve technical problems such as the difficulty in fault diagnosis of rolling bearings and the high rate of misjudgment, the research on the fault diagnosis system of bearings has become more and more meaningful.

In the existing various bearing fault diagnosis technologies, the analysis of vibration signals is still the main method. The signal frequency domain representation based on Fourier transform reveals the inner connection between the time function and the spectral function, which plays an extremely important role in the traditional analysis and processing of stationary signals, but the Fourier transform method is used to extract the signal When using the frequency spectrum, it is necessary to use all the time-domain information of the signal, which is an overall transformation and lacks the time-domain positioning function. Wavelet transform is a new transformation analysis method in recent years. It inherits and develops the idea of short-time Fourier transform localization, and at the same time overcomes the shortcomings of the window size not changing with frequency. Wavelet transform can better observe the localization of the signal characteristics, the time and frequency information of the signal can be observed at the same time, which is beyond the reach of the Fourier transform. However, the use of wavelet transform to suppress noise often causes oscillation effects when dealing with low signal-to-noise ratio signals. At the same time, it is necessary to manually select a suitable wavelet when performing wavelet transform, which greatly increases the time required for the processing of practical engineering problems. The EMD (Empirical Mode Decomposition Method) proposed by Dr. Huang E, a member of the American Academy of Engineering, has been proven to be an adaptive data processing and mining method, which is very suitable for the processing of nonlinear and non-stationary time series without presetting any basis functions. , which is essentially a smoothing of the data sequence or signal. The purpose of the EMD algorithm is to decompose the signal with poor performance into a group of intrinsic mode functions with better performance, and each component decomposed contains local characteristic signals of different time scales of the original signal. Select a suitable component, and then perform Hilbert transform to obtain a time-spectrogram, and obtain a physically meaningful frequency. However, in the working environment, how to correctly choose the amplitude of the added noise still needs further research. The frequent occurrence of modal aliasing is also one of the main shortcomings of EMD. The modal aliasing is caused by signal interruption, and interruption is an indefinite form of disturbance. Signals are often encountered in actual processing. Disruptions can lead to confusing time-frequency distributions, which in turn destroy the physical meaning of the IMF.

Contents of the invention

The technical problem to be solved by the embodiments of the present invention is to provide a nuclear regression decomposition method and system for rolling bearing fault diagnosis, and use the nuclear regression decomposition method to process and analyze vibration signals, extract bearing-related features and identify them.

To achieve the above object, the technical solution of the present invention comprises the following steps:

(1) Data collection, using an acceleration sensor to collect the vibration signal data of the rolling bearing to be tested, the acceleration sensor is installed on the bearing seat of the bearing to be tested, the output end of the acceleration sensor is connected with a multi-channel data acquisition analyzer, and the multi-channel data acquisition analyzer Save the extracted data and transfer it to the computer;

(2) adopt nuclear regression decomposition method to carry out signal processing to the real-time data that accelerometer measures, nuclear regression decomposition method:

(2.1) Firstly, the data obtained by the multi-channel data acquisition analyzer is used as the original signal to carry out the first kernel regression process with the Gaussian kernel function to obtain a residual variable, and the used kernel regression decomposition formula is:

In the formula: f _∑,1 (t) is the original signal, f _∑,1 (t _i ) is the kernel function center, λ ₁ is the bandwidth parameter, f _∑,2 (t) is

The new signal obtained after kernel regression transformation, K ₁ is a Gaussian kernel function;

(2.2) Then subtract the obtained residual variable from the original signal to obtain a new component expressed as the first residual component, and then perform kernel regression formula processing on f _∑,2 (t), and so on to continuously obtain new Residual components, but all decompositions are not endless decompositions, stop decomposition when we meet our given standard deviation standard, the formula is:

In the formula: T is the length of data, and C _j (t) is the component obtained with nuclear regression; When satisfying the threshold value set in the above formula, the nuclear regression process stops and carries out next step analysis;

Then, perform soft threshold processing on some redundant information contained in each new residual component caused by incomplete processing, and further denoise the obtained components. The threshold formula is:

θ _j = MAD(C _j (t))/0.6745 (3)

In the formula: MAD is the mean absolute deviation;

(2.3) The last step of kernel regression is to reconstruct all the residual components obtained, and accumulate all the residual components to obtain a new signal for data analysis and fine fault diagnosis;

(3) Finally, after all the components are superimposed to obtain a new signal, the Hilbert envelope spectrum analysis is performed, and the rolling bearing fault information is obtained in the time-frequency conversion.

It is further set that the threshold value of the kernel regression processing stop set by the formula (2) in the step (2.2) is 0.2.

The present invention also provides a rolling bearing fault diagnosis system based on the nuclear regression decomposition method, which includes a transmission shaft for installing the rolling bearing to be tested, a power unit for driving the rotation of the transmission shaft, and is arranged on the bearing seat of the bearing to be tested for The acceleration sensor for collecting the acceleration data of the roller bearing to be tested, the multi-channel data acquisition analyzer and the computer, the output end of the acceleration sensor is connected to the multi-channel data acquisition analyzer, and the multi-channel data acquisition analyzer saves the extracted data and transmits them to the computer;

Adopt kernel regression decomposition method to carry out signal processing to the real-time data that acceleration sensor records in described computer, kernel regression decomposition method:

First, the data obtained by the multi-channel data acquisition analyzer is used as the original signal to perform the first kernel regression processing with the Gaussian kernel function to obtain a residual variable. The formula for kernel regression decomposition used is:

In the formula: f _∑,1 (t) is the original signal, f _∑,1 (t _i ) is the kernel function center, λ ₁ is the bandwidth parameter, f _∑,2 (t) is the new signal obtained after kernel regression transformation, K ₁ is a Gaussian kernel function;

Then subtract the obtained residual variable from the original signal to obtain a new component expressed as the first residual component, and then perform kernel regression formula processing on f _∑,2 (t), and so on to continuously obtain new residual components, But all the decompositions are not endless decompositions. When the standard deviation standard we give is met, the decomposition is stopped. The formula is:

In the formula: T is the length of data, and C _j (t) is the component obtained with nuclear regression; When satisfying the threshold value set in the above formula, the nuclear regression process stops and carries out next step analysis;

Then, perform soft threshold processing on some redundant information contained in each new residual component caused by incomplete processing, and further denoise the obtained components. The threshold formula is:

θ _j = MAD(C _j (t))/0.6745 (3)

In the formula: MAD is the mean absolute deviation;

The last step of kernel regression is to reconstruct all the residual components obtained, and obtain a new signal by accumulating all the residual components for data analysis and fine fault diagnosis;

Finally, after all the components are superimposed to obtain a new signal, the Hilbert envelope spectrum analysis is carried out, and the fault information of the rolling bearing is obtained in the time-frequency conversion.

It is further provided that the power unit includes a motor, the output shaft of the motor is connected to the transmission shaft through a belt transmission unit, and the transmission shaft is installed with a rolling bearing to be tested through a coupling.

The kernel regression method used in the present invention is a new digital signal processing method. Based on the principle of EMD, the signal will be decomposed into several scales by kernel regression technology at first, and each scale contains certain characteristic information, but each scale also contains certain characteristic information. Contains some useless information. Therefore, this algorithm is optimized by combining the soft threshold method and the standard deviation standard to achieve the effect of denoising. Finally, the Hilbert envelope spectrum method is used to convert the signal to time-frequency, and the result is obtained after processing the new signal. And judge the fault information.

Through practical application, the present invention has a good effect in processing rolling bearing fault signals, and the method is more suitable for denoising and fault detection in mechanical systems.

The beneficial effects of the invention are: the invention has the advantages of good transmission performance, high speed, simple operation, timely detection of machine rolling bearing faults, and good engineering application effect.

Description of drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only These are some embodiments of the present invention. For those of ordinary skill in the art, obtaining other drawings based on these drawings still belongs to the scope of the present invention without any creative effort.

Fig. 1 is the fault diagnosis diagram of kernel regression decomposition method of the present invention;

Fig. 2 is the kernel regression method roadmap of the present invention;

Fig. 3 is the processing result of the bearing outer ring fault signal of the present invention;

Fig. 4 is the signal processing result of the complex fault of the bearing of the present invention.

detailed description

In order to make the object, technical solution and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the accompanying drawings.

The terms of direction and position mentioned in the present invention, such as "up", "down", "front", "back", "left", "right", "inside", "outside", "top", "bottom" ", "side", etc., are only referring to the direction or position of the drawings. Therefore, the terms used in direction and position are used to explain and understand the present invention, but not to limit the protection scope of the present invention.

As shown in Fig. 1 to Fig. 4, in the embodiment of the present invention, comprise motor, belt transmission shaft, shaft coupling, faulty bearing, acceleration sensor, multi-channel data acquisition analyzer, PC machine, the faulty bearing described in Fig. 1 passes through The motor, belt drive shaft and coupling are connected to the drive shaft, the acceleration sensor is fixed on the bearing seat of the test bearing, the output end of the acceleration sensor is connected to the multi-channel data acquisition analyzer, and the analyzer transmits the extracted data to the computer after saving In this embodiment, the computer adopts a traditional X86 or X64 PC, and the PC combines kernel regression decomposition technology to analyze and process the signal data, so as to realize accurate detection of the running state of the rolling bearing. The operation steps are as follows:

(1) at first select data collector, the present invention adopts acceleration sensor to collect data, and this acceleration sensor is installed on the bearing block of bearing to be tested, and sampling frequency is generally 25600Hz, according to the concrete parameter setting of sensor, on Fig. 2 experimental bench After the measured data results are transmitted to the PC in real time.

(2) Secondly, the present invention will use the kernel regression decomposition method to carry out signal processing to the real-time data measured by the acceleration sensor. Kernel regression decomposition method:

First, the data obtained by the multi-channel data acquisition analyzer is used as the original signal to perform the first kernel regression processing with the Gaussian kernel function to obtain a residual variable. The formula for kernel regression decomposition used is:

In the formula: f _∑,1 (t) is the original signal, f _∑,1 (t _i ) is the kernel function center, λ ₁ is the bandwidth parameter, f _∑,2 (t) is the new signal obtained after kernel regression transformation, K ₁ is a Gaussian kernel function.

Then subtract the obtained residual variable from the original signal to obtain a new component, which is expressed as the first residual component, and then perform kernel regression formula processing on f _Σ,2 (t), and so on to continuously obtain new residual components. But all the decompositions are not endless decompositions, and the decomposition stops when the standard deviation standard we give is met. Its formula is:

In the formula: T is the length of the data, and C _j (t) is the component obtained by kernel regression. When the threshold set in the above formula is met, generally we set it to 0.2, the kernel regression processing stops for the next step of analysis.

In addition, in the above processing process, the processing may not be complete, and each new residual component contains some redundant information. Therefore, the present invention uses the method of EMD for reference to perform soft threshold processing, and further eliminates the obtained components. noise processing. Its threshold formula is:

θ _j = MAD(C _j (t))/0.6745 (3)

In the formula: MAD is the mean absolute deviation.

The last step of kernel regression is to reconstruct all the residual components obtained, and accumulate all the residual components to obtain a new signal for data analysis and fine fault diagnosis.

(3) Finally, after all the components are superimposed to obtain a new signal, the Hilbert envelope spectrum analysis is performed, and the bearing fault information is obtained in the time-frequency conversion. The specific operation process is shown in Figure 2.

As shown in Figure 1, the kernel regression decomposition technology bearing fault diagnosis system, the fault information is collected by the acceleration sensor.

As shown in Figure 1, the acceleration sensor is installed on the faulty bearing, placed on the bearing seat of the test bearing, and the vibration signal in the radial direction is sampled.

As shown in Figure 1, the model of the multi-channel data acquisition analyzer is AVANT-MI-7016, which has 16 channels, and each channel has the functions of signal acquisition, extraction, filtering, and signal source output, which makes the system more efficient in terms of hardware. It is reliable and simplifies many intermediate steps of signal processing, practical and convenient.

Referring to Fig. 1, Fig. 1 is a structural block diagram of a specific embodiment of the rolling bearing fault diagnosis system provided by the present invention.

The rolling bearing fault diagnosis system is used to detect the fault phenomenon of the rolling bearing, and upload the fault vibration signal to the PC. In the specific scheme, the motor drives the belt transmission, and the coupling and transmission shaft provide the rotational speed to the simulated test bench, and collect parameter signals through the acceleration sensor on the faulty bearing, and process these collected parameter signals through the kernel regression method, so that The condition of the equipment is judged, and the description of the condition of the equipment is the fault phenomenon, such as the fault phenomenon of the inner ring, outer ring, and rolling body of the bearing. The specific calculation method can be calculated by the following formula:

Outer ring fault formula:

Inner ring fault formula:

Rolling element failure formula:

In the formula: f _r is the rotation frequency, n is the number of bearing rolling elements, φ is the contact angle in the radial direction, d is the average diameter of the rolling elements, and D is the average diameter of the bearing.

Example case 1:

As shown in Figure 3, it is the result map processed by the rolling bearing fault diagnosis system. According to the existing bearing information, the fault frequency of the outer ring of the bearing in this experiment can be calculated as 91.15Hz, and the fault can be quickly identified from the graph. The frequency is 87.5Hz, which is close to the theoretical value, and 29.8Hz is the rotation frequency of the motor. It can be determined that the fault type of this rolling bearing is the fault of the outer ring of the bearing.

Implementation case 2:

On the basis of Example 1, this implementation case 2 will further deal with more complex bearing faults. By calculating the faulty bearing in this experiment is a composite fault including bearing inner ring, outer ring and rolling element faults. Through the calculation of the data of the test bearing, the fault frequency of the inner ring is 197.05Hz, the fault frequency of the outer ring is 121.51Hz, and the fault frequency of the rolling element is 79.25Hz. From Figure 4, it can be concluded that the frequency is 120Hz matching the outer ring fault, 160Hz is consistent with the double frequency of the rolling element fault, and 202.5Hz is basically consistent with the inner ring fault. Thereby it is further illustrated that the present invention has a better treatment effect and is worthy of popularization and application.

Those of ordinary skill in the art can understand that all or part of the steps in the method of the above-mentioned embodiments can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium, and the storage Media such as ROM/RAM, magnetic disk, optical disk, etc.

The above disclosures are only preferred embodiments of the present invention, and certainly cannot limit the scope of rights of the present invention. Therefore, equivalent changes made according to the claims of the present invention still fall within the scope of the present invention.

Claims (4)

1. the kernel regression decomposition method of a kind of rolling bearing fault diagnosis, it is characterised in that comprise the following steps：

(1) data acquisition, the vibration signal data of rolling bearing to be measured, the acceleration transducer are gathered using acceleration transducer On the bearing block of bearing to be measured, acceleration transducer output end is connected with multi-channel data acquisition analyzer, multichannel Data collection and analysis instrument is by the data after extraction by being sent to computer after preservation；

(2) signal transacting, kernel regression point are carried out to the real time data measured by acceleration transducer using kernel regression decomposition method Solution method：

(2.1) data for first obtaining multi-channel data acquisition analyzer carry out first as primary signal with gaussian kernel function Secondary kernel regression processing obtains a remaining variable, and the formula that kernel regression used is decomposed is：

<mrow> <msub> <mi>f</mi> <mrow> <mi>&amp;Sigma;</mi> <mo>,</mo> <mn>2</mn> </mrow> </msub> <mrow> <mo>(</mo> <mi>t</mi> <mo>)</mo> </mrow> <mo>=</mo> <msub> <mi>K</mi> <mn>1</mn> </msub> <mrow> <mo>(</mo> <msub> <mi>f</mi> <mrow> <mi>&amp;Sigma;</mi> <mo>,</mo> <mn>1</mn> </mrow> </msub> <mo>(</mo> <mi>t</mi> <mo>)</mo> <mo>-</mo> <msub> <mi>f</mi> <mrow> <mi>&amp;Sigma;</mi> <mo>,</mo> <mn>1</mn> </mrow> </msub> <mo>(</mo> <msub> <mi>t</mi> <mi>i</mi> </msub> <mo>)</mo> <mo>)</mo> </mrow> <mo>=</mo> <msup> <mi>e</mi> <mrow> <mo>-</mo> <mfrac> <mn>1</mn> <mrow> <mn>2</mn> <msup> <msub> <mi>&amp;lambda;</mi> <mn>1</mn> </msub> <mn>2</mn> </msup> </mrow> </mfrac> <msup> <mrow> <mo>(</mo> <msub> <mi>f</mi> <mrow> <mi>&amp;Sigma;</mi> <mo>,</mo> <mn>1</mn> </mrow> </msub> <mo>(</mo> <mi>t</mi> <mo>)</mo> <mo>-</mo> <msub> <mi>f</mi> <mrow> <mi>&amp;Sigma;</mi> <mo>,</mo> <mn>1</mn> </mrow> </msub> <mo>(</mo> <msub> <mi>t</mi> <mi>i</mi> </msub> <mo>)</mo> <mo>)</mo> </mrow> <mn>2</mn> </msup> </mrow> </msup> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>1</mn> <mo>)</mo> </mrow> </mrow>

In formula：f_∑,1(t) it is primary signal, f_∑,1(t_i) it is kernel function center, λ₁For bandwidth parameter, f_∑,2(t) converted for kernel regression The new signal obtained afterwards, K₁For gaussian kernel function；

(2.2) and then by the remaining variable of gained and original signal subtract each other that to obtain a new representation in components be first remnants points Amount, then to f_∑,2(t) kernel regression formula manipulation is carried out again, and new residual components, but all decomposition are continuously available by that analogy It is not endless decomposition, stops decomposing when meeting the standard deviation standard that we give, its formula is:

<mrow> <mi>S</mi> <mi>D</mi> <mrow> <mo>(</mo> <mi>k</mi> <mo>)</mo> </mrow> <mo>=</mo> <munderover> <mo>&amp;Sigma;</mo> <mrow> <mi>t</mi> <mo>=</mo> <mn>0</mn> </mrow> <mi>T</mi> </munderover> <mo>&amp;lsqb;</mo> <mfrac> <mrow> <mo>|</mo> <msub> <mi>C</mi> <mrow> <mi>j</mi> <mo>-</mo> <mn>1</mn> </mrow> </msub> <mrow> <mo>(</mo> <mi>t</mi> <mo>)</mo> </mrow> <mo>-</mo> <msub> <mi>C</mi> <mi>j</mi> </msub> <mrow> <mo>(</mo> <mi>t</mi> <mo>)</mo> </mrow> <msup> <mo>|</mo> <mn>2</mn> </msup> </mrow> <mrow> <msub> <msup> <mi>C</mi> <mn>2</mn> </msup> <mrow> <mi>j</mi> <mo>-</mo> <mn>1</mn> </mrow> </msub> <mrow> <mo>(</mo> <mi>t</mi> <mo>)</mo> </mrow> </mrow> </mfrac> <mo>&amp;rsqb;</mo> <mo>&amp;le;</mo> <mn>0.2</mn> <mo>-</mo> <mn>0.3</mn> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>2</mn> <mo>)</mo> </mrow> </mrow>

In formula:T is the length of data, C_j(t) component to be obtained with kernel regression；When meeting the threshold value that is set in above formula, kernel regression Processing stops carrying out next step analysis；

Then, for without processing it is complete caused by included in each new residual components some unnecessary information carry out Soft-threshold processing, further carries out denoising Processing by the component of gained, and its threshold formula is：

θ_j=MAD (C_j(t))/0.6745 (3)

In formula：MAD is mean absolute deviation；

(2.3) final step of kernel regression is exactly that all obtained residual components are reconstructed, by by all remnants points Amount carries out obtaining new signal to carry out data analysis and fine fault diagnosis after adding up；

(3) finally all components are superimposed and obtain carrying out Hilbert envelope analysis of spectrum after new signal, in time-frequency convert In draw rolling bearing fault information.

2. a kind of kernel regression decomposition method of rolling bearing fault diagnosis according to claim 1, it is characterised in that：It is described The threshold value that kernel regression processing in step (2.2) set by formula (2) stops is 0.2.

3. a kind of Diagnosing System for Detecting of Antifriction Bearings based on kernel regression decomposition method, it is characterised in that：Include for installing The power transmission shaft of rolling bearing to be measured, for driving power unit positioned at drive axis, being arranged at the bearing block of bearing to be measured Upper acceleration transducer and multi-channel data acquisition analyzer and the calculating for being used to gather roller bearing acceleration information to be measured Machine, acceleration transducer output end is connected with multi-channel data acquisition analyzer, after multi-channel data acquisition analyzer will be extracted Data by being sent to computer after preservation；

Signal transacting is carried out to the real time data measured by acceleration transducer using kernel regression decomposition method in the computer, Kernel regression decomposition method：

The data that multi-channel data acquisition analyzer is obtained first carry out first time core as primary signal with gaussian kernel function Recurrence processing obtains a remaining variable, and the formula that kernel regression used is decomposed is：

<mrow> <msub> <mi>f</mi> <mrow> <mi>&amp;Sigma;</mi> <mo>,</mo> <mn>2</mn> </mrow> </msub> <mrow> <mo>(</mo> <mi>t</mi> <mo>)</mo> </mrow> <mo>=</mo> <msub> <mi>K</mi> <mn>1</mn> </msub> <mrow> <mo>(</mo> <mrow> <msub> <mi>f</mi> <mrow> <mi>&amp;Sigma;</mi> <mo>,</mo> <mn>1</mn> </mrow> </msub> <mrow> <mo>(</mo> <mi>t</mi> <mo>)</mo> </mrow> <mo>-</mo> <msub> <mi>f</mi> <mrow> <mi>&amp;Sigma;</mi> <mo>,</mo> <mn>1</mn> </mrow> </msub> <mrow> <mo>(</mo> <msub> <mi>t</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> </mrow> <mo>)</mo> </mrow> <mo>=</mo> <msup> <mi>e</mi> <mrow> <mo>-</mo> <mfrac> <mn>1</mn> <mrow> <mn>2</mn> <msup> <msub> <mi>&amp;lambda;</mi> <mn>1</mn> </msub> <mn>2</mn> </msup> </mrow> </mfrac> <msup> <mrow> <mo>(</mo> <mrow> <msub> <mi>f</mi> <mrow> <mi>&amp;Sigma;</mi> <mo>,</mo> <mn>1</mn> </mrow> </msub> <mrow> <mo>(</mo> <mi>t</mi> <mo>)</mo> </mrow> <mo>-</mo> <msub> <mi>f</mi> <mrow> <mi>&amp;Sigma;</mi> <mo>,</mo> <mn>1</mn> </mrow> </msub> <mrow> <mo>(</mo> <msub> <mi>t</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> </mrow> <mo>)</mo> </mrow> <mn>2</mn> </msup> </mrow> </msup> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>1</mn> <mo>)</mo> </mrow> </mrow>

In formula：f_∑,1(t) it is primary signal, f_∑,1(t_i) it is kernel function center, λ₁For bandwidth parameter, f_∑,2(t) converted for kernel regression The new signal obtained afterwards, K₁For gaussian kernel function；

Then the remaining variable of gained and original signal subtracted each other to obtain a new representation in components be first residual components, then To f_∑,2(t) kernel regression formula manipulation is carried out again, and new residual components are continuously available by that analogy, but all decomposition are not It is endless to decompose, stop decomposing when meeting the standard deviation standard that we give, its formula is:

<mrow> <mi>S</mi> <mi>D</mi> <mrow> <mo>(</mo> <mi>k</mi> <mo>)</mo> </mrow> <mo>=</mo> <munderover> <mo>&amp;Sigma;</mo> <mrow> <mi>t</mi> <mo>=</mo> <mn>0</mn> </mrow> <mi>T</mi> </munderover> <mo>&amp;lsqb;</mo> <mfrac> <mrow> <mo>|</mo> <msub> <mi>C</mi> <mrow> <mi>j</mi> <mo>-</mo> <mn>1</mn> </mrow> </msub> <mrow> <mo>(</mo> <mi>t</mi> <mo>)</mo> </mrow> <mo>-</mo> <msub> <mi>C</mi> <mi>j</mi> </msub> <mrow> <mo>(</mo> <mi>t</mi> <mo>)</mo> </mrow> <msup> <mo>|</mo> <mn>2</mn> </msup> </mrow> <mrow> <msub> <msup> <mi>C</mi> <mn>2</mn> </msup> <mrow> <mi>j</mi> <mo>-</mo> <mn>1</mn> </mrow> </msub> <mrow> <mo>(</mo> <mi>t</mi> <mo>)</mo> </mrow> </mrow> </mfrac> <mo>&amp;rsqb;</mo> <mo>&amp;le;</mo> <mn>0.2</mn> <mo>-</mo> <mn>0.3</mn> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>2</mn> <mo>)</mo> </mrow> </mrow>

In formula:T is the length of data, C_j(t) component to be obtained with kernel regression；When meeting the threshold value that is set in above formula, kernel regression Processing stops carrying out next step analysis；

Then, for without processing it is complete caused by included in each new residual components some unnecessary information carry out Soft-threshold processing, further carries out denoising Processing by the component of gained, and its threshold formula is：

θ_j=MAD (C_j(t))/0.6745 (3)

In formula：MAD is mean absolute deviation；

The final step of kernel regression is exactly that all obtained residual components are reconstructed, by the way that all residual components are carried out Obtain new signal to carry out data analysis and fine fault diagnosis after cumulative；

Finally all component superpositions are obtained to carry out Hilbert envelope analysis of spectrum after new signal, in time-frequency convert Go out rolling bearing fault information.

4. a kind of Diagnosing System for Detecting of Antifriction Bearings based on kernel regression decomposition method according to claim 3, its feature It is：Power unit includes motor, and the output shaft of motor is connected by belt transmission unit with transmission shaft driven.