A method for mass detection of rolling bearings

Abstract

The application belongs to the field of equipment quality detection, and proposes a quality detection method for rolling bearings. The ratio of the sum of the amplitude of the fault frequency and its multiple frequency of each component of the rolling bearing to the amplitude of the rotation frequency is selected as the parameter index for judging the quality of each component of the bearing, which is used for positioning the unqualified components of the bearing, and the failure rate is reduced through the development of a calibration scheme. The application detects the quality of the rolling bearing based on the vibration signal, and can determine the quality state of the rolling bearing by identifying the distribution of the modulated vibration and noise generated by the processing corrugation and accidental damage of the inner and outer slides and rolling bodies in the signal spectrum. Compared with the current detection standard, the unqualified components of the bearing can be accurately positioned, and a strong direction is provided for the optimization direction of the production process. In addition, the application can be used for process comparison of each workshop of the enterprise by establishing a bearing component damage degree model through a hidden Markov model, which can provide a strong direction for the optimization of the production process.

Description

Technical Field

[0001] This invention relates to the field of equipment quality inspection, and more particularly to a quality inspection method for rolling bearings. Background Technology

[0002] Rolling bearings, often called the "joints of industry," are crucial supports for shafts and other rotating mechanical components, finding widespread application in daily life, industrial production, and national defense. While their structure appears simple, rolling bearings are actually highly precise, requiring complex manufacturing processes. Due to technological limitations, individual variations in bearings are unavoidable. Ensuring bearing quality is paramount to guaranteeing smooth and safe production. Therefore, before leaving the factory, bearings undergo quality inspection to identify defective units, pinpoint production process flaws, and optimize the entire manufacturing process.

[0003] The main standards for inspecting finished rolling bearings include national standards, industry standards, and enterprise standards. The inspection items are numerous, primarily including surface cracks, hardness, clearance, dimensional accuracy, rotational accuracy, surface waviness, and vibration. Vibration testing is a crucial part of the entire inspection process, as vibration signals contain rich information about the bearing's condition and can accurately reflect its health status. Currently, my country has two parallel standards for vibration signal-based bearing inspection: the speed standard proposed in JB / T 10187-2000 and the acceleration standard proposed in JB / T 7047-2006, commonly known as the V standard and Z standard. Both standards involve collecting bearing vibration signals for spectral analysis; if the vibration velocity or acceleration value of a single bearing exceeds a specified peak value in different frequency ranges, it is considered unqualified. Such inspection standards are quite crude, only able to detect the overall quality of a single rolling bearing, unable to inspect the quality of the core components of the bearing—the inner and outer rings, rolling elements, and support—and unable to accurately pinpoint the location of bearing defects. In addition, many companies have proposed their own testing methods and standards based on national standards, but most of these have low applicability and cannot be widely adopted. Therefore, the bearing testing industry urgently needs a new, efficient testing method that can quantify the quality of the various components that make up a bearing.

[0004] To address the aforementioned problems, this invention proposes a quality inspection method for rolling bearings. Based on vibration and velocity signals, it determines the quality of the inner and outer raceways and rolling elements of the rolling bearing. Furthermore, it locates quality issues in various components of the rolling bearing by analyzing the modulation noise generated on the bearing surface due to machining roughness and damage. The implementation process of this invention can be used for the formulation and calibration of industrial production inspection standards, providing strong guidance for optimizing production processes. Summary of the Invention

[0005] To address the current problem that pre-shipment quality inspection standards for bearings cannot determine the presence and extent of damage to core bearing components, this invention provides a quality inspection method for rolling bearings. This method detects the quality of each bearing component based on vibration noise and velocity signals. It assesses the quality of each component by identifying modulation noise generated by surface roughness or damage, extending bearing quality inspection from individual component testing to component-level testing. Furthermore, it models the surface damage level of each bearing component using a Hidden Markov Model (HMM), which can serve as a reference for comparing processing technologies across different workshops within an enterprise, and can provide assistance in optimizing the production processes of each bearing component.

[0006] This invention relies on the following parameters for quality detection of the inner and outer rings and rolling elements of a bearing: spindle rotation frequency, fault characteristic frequencies of each component, and their modulation noise. The spindle rotation frequency is the main component of the vibration signal, exhibiting a large amplitude in the frequency spectrum, and its energy proportion is significantly higher than that of various fault frequencies. Fault characteristic frequencies are low-frequency vibration components generated by repeated impacts between damaged points on the surface of various bearing components and the surfaces of other components they contact during operation. Their periods are regular, but the frequencies of the noise generated differ when damage occurs on the inner and outer raceways and the rolling element surfaces.

[0007] With the outer ring fixed and the inner ring rotating, the formulas for the spindle speed and the theoretical fault characteristic frequency caused by damage at only a certain point are as follows:

[0008]

[0009]

[0010]

[0011]

[0012] In the formula, f r It is the rotational frequency of the shaft. N The rotational speed of the shaft. f ic It is the frequency with which a single rolling element passes through a damage point on the inner raceway. f oc It is the frequency with which a single rolling element passes a point of damage on the outer raceway. f bc It is the frequency with which a single damage point on the surface of a rolling element passes between the inner and outer rings. d It is the diameter of the rolling element. D It is the diameter of the circle containing the center of the bearing rolling elements. It is the angle between the direction of the force on the rolling element and the perpendicular lines of the inner and outer raceways.

[0013] This invention selects the ratio of the sum of the amplitudes of the failure frequencies and their harmonics of each component of a rolling bearing to the amplitude of the rotational frequency as a parameter for evaluating the quality of each component. Theoretically, the larger the surface defects of each component of the bearing, the larger this parameter will be.

[0014] The technical solution adopted in this invention is: a quality inspection method for rolling bearings, specifically including the following steps:

[0015] Step 1: Select healthy rolling bearings as samples for calibrating and collecting vibration velocity values ​​of the rolling bearings to obtain a test data sequence;

[0016] Step 2: Based on the original waveform and envelope spectrum of the vibration signal in the detection data sequence, identify the spindle speed and vibration amplitude of the bearing, and calculate the theoretical fault characteristic frequency of the rolling bearing. Although the spindle speed can be measured by installing a speed sensor, in many complex industrial sites, it is not possible to directly install a speed sensor to measure the speed of the spindle where the bearing is located. It is necessary to analyze the vibration signal. Therefore, this invention provides a method for identifying the synchronous speed of the bearing based on the original waveform and envelope spectrum of the vibration signal, thereby obtaining the theoretical fault characteristic frequency of the bearing's outer raceway.

[0017] Step 3: The vibration signal in the detection data sequence is decomposed using Variational Mode Decomposition (VMD) to find the theoretical fault characteristic frequency and its corresponding harmonic amplitude from each modal component. Combined with the spindle speed obtained in Step 2, the sum of the amplitudes of the theoretical fault characteristic frequencies and their harmonics as the rolling bearing gradually deteriorates from healthy to damaged states is used as a parameter to evaluate the quality of each bearing component. Because the vibration components are complex when the rolling bearing surface experiences significant wear, including harmonic components, noise, and impact response, directly identifying fault characteristics from the original vibration spectrum is quite difficult. Therefore, it is necessary to first reveal the characteristics of the submerged fault characteristic frequencies. This invention uses Variational Mode Decomposition (VMD) to decompose the original signal, weakening the influence of various interference signals, and finding the target frequency from each modal component.

[0018] Step 4: When the bearing is healthy or slightly worn, the rotational frequency is the main component of the spectrum, and the energy is higher than the fault characteristic frequency. As the damage worsens, the fault characteristic frequency will gradually increase. Select parameters that exceed the set values ​​as initial parameters and periodically check and adjust the parameters obtained in the above steps to reduce the error rate.

[0019] Through steps one through four described above, this invention proposes a process for enterprises to formulate quality inspection standards for various bearing components and a method for periodically revising these standards, using these standards to determine whether each bearing component is qualified. However, such standards cannot determine the degree of non-compliance, which necessitates enterprises to establish damage models for each bearing component based on substantial data support for assessment.

[0020] Step 5: Establish a Hidden Markov Model (HMM) for bearing damage, classify the quality of each bearing component, and compare the quality of the production process.

[0021] The healthy rolling bearings were screened according to the JB / T 10187-2000 national bearing testing standard. Testing bearings and vibration velocity sensors were installed on the bearing housing of the bearing quality testing bench to collect bearing operation data. The bearing quality testing bench included a variable frequency drive motor 1, rolling bearings, a bearing housing 5, a coupling 2, a horizontal / vertical velocity sensor 9, and a radial force loading device 4. The variable frequency drive motor 1 sequentially drove the comparison bearing 6 and the bearing under test 7 through the coupling 2 and was connected to a magnetic powder brake 3. The radial force loading device 4 was installed between the comparison bearing 6 and the bearing under test 7. A temperature sensor 8 and a horizontal / vertical velocity sensor 9 were installed on the bearing under test 7. The outer ring of the rolling bearing was fixed, and the inner ring rotated at a constant speed under a fixed load. The horizontal / vertical velocity sensor 9 was installed at half the width of the outer cylindrical surface of the bearing outer ring, with the measurement direction along the bearing radial direction and perpendicular to the bearing axis.

[0022] In step two, envelope spectrum analysis is first performed on the vibration signal in the detection data sequence. Within the effective frequency band of the envelope spectrum, local amplitudes are searched from low frequencies. The values ​​of these local amplitudes are higher than the noise in the surrounding frequency bands, and the frequency corresponding to the highest amplitude is marked as [the highest amplitude value]. f rt This frequency is tentatively set as the target switching frequency and needs to be calibrated; select the target switching frequency. f rt Using the center frequency as the center frequency and the rotational speed error as the bandwidth dimension, the spectrum of the original vibration signal is bandpass filtered. The bandpass filtering interval is... ,in, f rt Given a provisional target rotational frequency and ΔR as the rotational speed error, the spindle rotational frequency time-domain signal is obtained, and the period of the spindle rotational frequency time-domain signal is calculated. T The rotational frequency of the spindle where the bearing is located is obtained. f r The relationship between the rotational speed and the rotational speed is ;

[0023] The rotational speed error is:

[0024]

[0025] In the formula, f c Where L is the sampling frequency and L is the number of FFT operation points;

[0026] The theoretical fault characteristic frequency caused by damage to the inner slide of the rolling bearing is obtained from the spindle speed. Theoretical fault characteristic frequency caused by damage to the outer slideway Theoretical fault characteristic frequency caused by damage to the rolling element surface They are respectively:

[0027]

[0028]

[0029]

[0030] in, n It is the vibration order. Z It is the number of rolling elements. d It is the diameter of the rolling element. D It is the diameter of the circle containing the center of the bearing rolling elements. α It is the angle between the direction of the force on the rolling element and the perpendicular lines of the inner and outer raceways.

[0031] In step three, the optimal solution of the VMD variational model is solved iteratively to obtain the intrinsic mode functions of each finite bandwidth, and the mode components are separated according to the center frequency of each mode component; the VMD variational model is as follows:

[0032]

[0033] in, x It is the original vibration signal. It is a variational mode component. It is the center frequency of the modal component; It is a unit impulse function; by setting a quadratic penalty factor and a Lagrange operator, the constrained variational model is transformed into an unconstrained variational model. After continuous iteration, the optimal solution of the variational model is searched using the alternating direction multiplier method, and finally the original vibration signal is decomposed into K IMF components. Since rolling bearings are all subjected to unilateral loads and have radial clearance, there can be relative displacement between the inner and outer rings in the radial direction. Depending on the location of the impact contact at the damage point, the amplitude of the vibration of the inner raceway and the rolling element surface may change periodically, resulting in amplitude modulation. The energy in the spectrum is expressed as local peaks and local peaks of their harmonics. Therefore, after performing a Hilbert transform on each modal component, the envelope spectrum is calculated. The theoretical fault characteristic frequency and the amplitude of its harmonics are found in the envelope spectrum of each modal component and summed to obtain the eigenvector. The denominator in the eigenvector E The numerator is the amplitude of the main axis rotation frequency, and the numerator is the summed amplitude; this eigenvector serves as the observation input value when building the subsequent model; the parameter indices are in ascending order, and indices greater than 0.75 are selected as the initial reference indices.

[0034] In step four, scrapped rolling bearings obtained during the maintenance process are used as samples to determine the effectiveness of parameter indicators. Scrapped bearings are divided into three types: bearings with obvious defects and scratches, bearings that have been scrapped due to their service life and have minor surface defects, and bearings that have been scrapped due to their service life but have good surface condition. Steps two and three are used to conduct quality inspections on the three types of scrapped bearings to obtain the pass rate of each type of scrapped bearing. The pass rate is compared with the set pass rate of each type of bearing, and then the magnitude of the parameter indicators is adjusted.

[0035] Step five specifically involves:

[0036] Step S5-1: Obtain the feature vector through step three. The Lloyd algorithm is used to perform scalar quantization on the feature vectors to form the observation sequence, and the initial probability is obtained by a random function. The initial state transition matrix A and the initial observation probability matrix B are used as parameters of the HMM model.

[0037] Step S5-2: Train the HMM model using the Baum-Welch algorithm; during training, the wear degree of each component of the bearing is divided into four states: normal state, inner ring damage state, outer ring damage state, and rolling element damage state, representing the four hidden states of the HMM model, denoted as . The observation sequence of the HMM model is the observation sequence provided in step S5-1; each set of feature vectors is vector quantized and used as the feature values ​​input to the HMM model for training the Baum-Welch algorithm; after training, the HMM model corresponding to the four states is obtained.

[0038] Step S5-3: When conducting quality inspection on bearings produced in the same batch, sample bearings are selected, vibration signals are collected, feature vectors are obtained, and these feature vectors are vector-quantized. The vector-quantized feature vectors are then input into the HMM model for each state, and each HMM model outputs a log-likelihood probability. The value represents the similarity between the observed sequence and the corresponding HMM model. The larger the value, the more similar the observed value is to the HMM model. The recognition algorithm used to find the optimal state sequence adopts the Viterbi algorithm. The state corresponding to the largest log-likelihood probability output is the damage state of the corresponding component of the bearing, thereby realizing the identification of the quality of each component of the bearing and judging the state of the bearing.

[0039] Step S5-4: Through step S5-3, the damage level of the same batch of bearings is obtained as an important criterion for evaluating the entire production process of the production workshop, or the quality of bearings of the same type produced in different workshops is compared to select the superior workshop as a reference object.

[0040] The set value for step four is selected as 0.75. Step four includes: based on the company's calibration cycle, for the scrapped bearings, sampling 40 sets each of bearings with obvious defects and scratches, bearings scrapped due to their service life with minor surface defects, and bearings scrapped due to their service life but in good surface condition. Using the reference index as the x-axis (range of the initial reference index obtained in step three and its surrounding small range) and the pass rate as the y-axis, three pass rate curves are plotted under different reference indices. The reference indices are appropriately adjusted so that the pass rate of the first type of bearing is closer to 0, the second type is as low as possible below 30%, and the third type is closer to 100%. Step four is repeated periodically to obtain the optimal judgment parameters.

[0041] This invention offers the following advantages: It provides a quality inspection method for rolling bearings, which uses vibration signals to detect bearing quality. By identifying the distribution of modulated vibrations and noise in the signal spectrum caused by machining ripples and accidental damage to the inner and outer slides and rolling elements, the quality status of the rolling bearing can be determined. Compared with current inspection standards, it can accurately locate defective bearing components, providing strong guidance for optimizing production processes. Furthermore, this invention relies entirely on vibration sensors installed in the industrial environment and the computing power of computers, making it unaffected by the industrial environment, highly operable, and widely applicable. Attached Figure Description

[0042] Figure 1 This is the overall flowchart of the present invention.

[0043] Figure 2 This is a structural diagram of the bearing quality testing bench described in this invention.

[0044] Figure 3 This is the process of establishing a hidden Markov model for the degree of damage to bearing components.

[0045] In the diagram: 1. Variable frequency drive motor; 2. Coupling; 3. Magnetic powder brake; 4. Radial force loading device; 5. Bearing housing; 6. Comparison bearing; 7. Bearing under test; 8. Temperature sensor; 9. Horizontal / vertical speed sensor. Detailed Implementation

[0046] The embodiments of the present invention are described in detail below with reference to the technical solutions and accompanying drawings.

[0047] Figure 1 The diagram shows a flow chart of a quality inspection method for rolling bearings proposed in this invention. The specific implementation steps are as follows:

[0048] Step S1: For bearings of the same type produced in the same batch, according to the JB / T 10187-2000 national bearing testing standard, 40 healthy bearings were selected as samples for calibration. Taking outer raceway damage as an example, the outer raceways of some samples were subjected to different degrees of damage processing. The bearings were then installed on a test bench and run for a period of time to obtain a stable operating state. The vibration velocity values ​​of the rolling bearings were collected, and the corresponding test data sequence was obtained. Note that the outer ring of the rolling bearing is fixed, while the inner ring rotates at a constant speed under a fixed load. The velocity sensor is installed at half the width of the outer cylindrical surface of the bearing outer ring, with the measurement direction along the radial direction of the bearing and perpendicular to the bearing axis.

[0049] Step S2: Identify the spindle rotation frequency and its vibration amplitude from the vibration signal, and obtain the theoretical fault characteristic frequency of the bearing outer raceway based on the bearing manufacturing process. The specific implementation is as follows:

[0050] Step S2-1: Perform envelope spectrum analysis on the vibration signal. Within the effective frequency band of the envelope spectrum, search for local amplitudes from the low frequencies. The values ​​of the local amplitudes should be significantly higher than the noise in the surrounding frequency bands. Mark the frequency corresponding to the highest amplitude as... f rt This frequency is tentatively set as the target rotational frequency. This is because during safe bearing operation, the rotational frequency signal is one of the main components, with high energy, manifesting as a high amplitude in the frequency domain, while the characteristic frequencies of the envelope spectrum are usually based on the rotational frequency and have high energy. According to the formula, the spindle speed where the bearing is located is: However, the rotational speed obtained at this time has a certain error, which is:

[0051]

[0052] In the formula, f c Where L is the sampling frequency and L is the number of FFT operation points.

[0053] Step S2-2: Increasing the number of FFT points can reduce the error, but it will make the frequency more dense, which is not ideal. Therefore, the result obtained in step S2-1 is selected. f rt Using the center frequency as the center frequency and the frequency shift error as the bandwidth dimension, bandpass filtering is performed on the original spectrum, i.e., the bandpass interval is... The spindle rotation frequency time-domain signal is obtained, which ideally resembles a sine wave. The period of the sine wave is then calculated. T From this, the rotational frequency of the spindle where the bearing is located can be obtained. f r Furthermore, the spindle speed can be obtained as

[0054]

[0055] Step S2-3: After obtaining the spindle speed, the theoretical fault characteristic frequencies caused by damage to the inner and outer slideways and rolling element surfaces of the rolling bearing are respectively...

[0056]

[0057]

[0058]

[0059] Step S3: The original signal is decomposed using Variational Mode Decomposition (VMD) to weaken the influence of various interference signals. The target fault characteristic frequency is identified from each modal component, and the corresponding frequency vibration amplitude is obtained. Combined with the spindle frequency obtained in Step S2, the ratio of the sum of the amplitudes of the theoretical fault characteristic frequencies and their harmonics (which gradually worsen from healthy to damaged) to the amplitude of the spindle frequency can be obtained. The specific implementation is as follows:

[0060] Step S3-1: It iteratively solves the optimal solution of the variational model to obtain the intrinsic mode functions of each finite bandwidth, and adaptively separates the modal components according to each center frequency. The variational model is as follows:

[0061]

[0062] in, x It is the original signal. It is a variational mode component. It is the center frequency of the modal component. By setting appropriate quadratic penalty factors and Lagrange operators, the constrained variational model is transformed into an unconstrained variational model. After continuous iteration, the optimal solution of the variational model is searched using the alternating direction multiplier method to find the optimal solution of the constraint factors. Finally, the original signal is decomposed into K IMF components.

[0063] Step S3-2: After performing a Hilbert transform on each mode, calculate the envelope spectrum. From the envelope spectrum of each modal component, find the fault characteristic frequency and the local peaks near their harmonics as the corresponding amplitudes, and sum them to obtain the eigenvector. This feature vector can be used as the observation input value for subsequent model building. The index is in ascending order, and the indexes that are exactly greater than 0.75 are selected as the initial reference indexes.

[0064] Step S4: Select initial parameter values ​​where the failure frequency is just higher than the rotational frequency (i.e., the parameter index exceeds 0.75). Regularly inspect and adjust the parameters obtained in the above steps to reduce the error rate. Use scrapped rolling bearings obtained during the inspection process as samples to determine the effectiveness of the parameter indicators. Scrapped bearings are roughly divided into three types: bearings with obvious defects and scratches, bearings scrapped due to their service life with minor surface defects, and bearings scrapped due to their service life but in good surface condition. Sample 40 sets of each type of bearing. Plot the pass rate curves under three different reference indicators, with the reference index as the x-axis (range of the aforementioned reference index and its surrounding small range) and the pass rate as the y-axis. Appropriately adjust the reference indicators so that the pass rate of the first type of bearing is closer to 0, the second type of bearing is as low as possible below 30%, and the third type of bearing is closer to 100%. Repeat step S4 periodically to obtain the optimal judgment parameter indicators.

[0065] Step S5: Using the sum of the amplitudes of the bearing failure frequencies and their harmonics at different damage levels, and the ratio of the amplitude of the rotational frequency, as input parameters, a Hidden Markov Model (HMM) for bearing damage is established to classify the quality of each component of the bearing. The specific implementation method is as follows:

[0066] Step S5-1: Obtain the eigenvector by combining the ratio of the bearing outer ring damage modulation frequency amplitude to the rotational frequency amplitude through steps S3 and S4. The Lloyd algorithm is used to perform scalar quantization on the above feature vectors to form the observation sequence, and the initial probability is obtained by a random function. The initial state transition matrix A and the initial observation probability matrix B are used to obtain the feature matrix.

[0067] Step S5-2: Train the model using the Baum-Welch algorithm. During training, the wear degree of each component of the bearing is roughly divided into four states, representing the four hidden states of the HMM model, denoted as... The observation sequence of the HMM model is the feature sequence provided above. The Baum-Welch algorithm requires discrete feature values ​​as input to train the HMM model, so vector quantization is performed on each set of feature vectors. During HMM training, the maximum likelihood estimation increases with the number of iterations, and the model eventually converges. After training, the HMM model corresponding to the four states is obtained.

[0068] Step S5-4: During quality inspection of a batch of produced bearings, bearing samples are randomly selected, vibration signals are collected, feature vectors are obtained, and these feature vectors are vector-quantized. The sequence of vector-quantized feature values ​​is then input into the HMM model for each state. Each model outputs a log-likelihood probability. The log-likelihood probability represents the degree of similarity between the observed value sequence and the corresponding HMM model. The larger the value, the more similar the observed value is to the HMM model. The identification algorithm adopts the Viterbi algorithm, and the state corresponding to the largest log-likelihood probability output is the damage state of the outer raceway of the bearing, thereby realizing the identification of the quality of the outer raceway of the bearing.

[0069] Thus, this invention completes the formulation of enterprise bearing inspection standards and the development of hidden Markov models for the degree of damage to various components. These can serve as important criteria for evaluating the entire production process of the production workshop, and can also be used to compare the quality of bearings of the same type produced in different workshops, selecting the better workshops as reference objects, which is of great help to the entire production process.

Claims

1. A quality inspection method for rolling bearings, characterized in that, The specific steps are as follows: Step 1: Select healthy rolling bearings as samples for calibrating and collecting vibration velocity values ​​of the rolling bearings to obtain a test data sequence; Step 2: Based on the original waveform and envelope spectrum of the vibration signal in the detection data sequence, identify the spindle speed and vibration amplitude of the rolling bearing, and calculate the theoretical fault characteristic frequency of the rolling bearing; Step 3: Use VMD processing to decompose the vibration signal in the detection data sequence, find the theoretical fault characteristic frequency and the amplitude corresponding to its harmonics from each modal component; combine with the spindle speed obtained in Step 2, obtain the ratio of the sum of the amplitudes of the theoretical fault characteristic frequency and its harmonics of the rolling bearing as it gradually worsens from healthy to damaged to the amplitude of the spindle speed as a parameter index for judging the quality evaluation of each component of the bearing. Step 4: Select parameters that exceed the set values ​​as initial parameters and periodically check and adjust the parameters obtained in the above steps to reduce the error rate. Step 5: Establish a Hidden Markov Model (HMM) for bearing damage, classify the quality of each bearing component, and compare the quality of the production process.

2. The quality inspection method for rolling bearings according to claim 1, characterized in that, The healthy rolling bearings are screened according to the national bearing testing standard JB / T 10187-2000; the bearings and vibration velocity sensors are installed on the bearing housing of the bearing quality testing bench to collect bearing operation data; the bearing quality testing bench includes a variable frequency drive motor (1), rolling bearings, bearing housing (5), coupling (2), horizontal / vertical velocity sensor (9) and radial force loading device (4); the variable frequency drive motor (1) drives the comparison bearing (6) and the bearing to be tested (7) in sequence through the coupling (2) and is connected to the magnetic powder brake (3); the radial force loading device (4) is installed between the comparison bearing (6) and the bearing to be tested (7); the temperature sensor (8) and the horizontal / vertical velocity sensor (9) are installed on the bearing to be tested (7); the outer ring of the rolling bearing is fixed, and the inner ring rotates at a constant speed under a fixed load; the horizontal / vertical velocity sensor (9) is installed at half the width of the outer cylindrical surface of the outer ring of the bearing, and the measurement direction is along the radial direction of the bearing and perpendicular to the axis of the bearing.

3. The quality inspection method for rolling bearings according to claim 1 or 2, characterized in that, In step two, envelope spectrum analysis is first performed on the vibration signal in the detection data sequence. Within the effective frequency band of the envelope spectrum, local amplitudes are searched from low frequencies. The values ​​of these local amplitudes are higher than the noise in the surrounding frequency bands, and the frequency corresponding to the highest amplitude is marked as f. rt This frequency is tentatively set as the target switching frequency; select the target switching frequency f rt Using the center frequency as the bandwidth, the spectrum of the original vibration signal is bandpass filtered, with the rotational speed error as the bandwidth dimension. The bandpass filtering interval is [f rt -ΔR / 60,f rt +ΔR / 60], where f rt Assuming a tentative target rotational frequency and ΔR as the rotational speed error, the spindle rotational frequency time-domain signal is obtained. The period T of the spindle rotational frequency time-domain signal is calculated, from which the rotational frequency f of the spindle where the bearing is located can be obtained. r Furthermore, the spindle speed can be obtained as The rotational speed error is: In the formula, f c Where L is the sampling frequency and L is the number of FFT operation points; The theoretical fault characteristic frequency f caused by damage to the inner slide of the rolling bearing is obtained from the spindle speed. is The theoretical fault characteristic frequency f caused by damage to the outer slideway os The theoretical fault characteristic frequency f caused by damage to the rolling element surface bs They are respectively: Where n is the vibration order, Z is the number of rolling elements, d is the diameter of the rolling elements, D is the diameter of the circle containing the center of the bearing rolling elements, and α is the angle between the direction of the force on the rolling elements and the perpendicular lines of the inner and outer raceways.

4. The quality inspection method for rolling bearings according to claim 3, characterized in that, In step three, the optimal solution of the VMD variational model is solved iteratively to obtain the intrinsic mode functions of each finite bandwidth, and the mode components are separated according to the center frequency of each mode component; the VMD variational model is as follows: Where x is the original vibration signal, {u k }={u1,…,u K } are variational mode components, {ω k }={ω1,…,ω K } represents the center frequency of the modal component; δ(t) is the unit impulse function; by setting a quadratic penalty factor and a Lagrange operator, the constrained variational model is transformed into an unconstrained variational model. After continuous iteration, the optimal solution of the variational model is searched using the alternating direction multiplier method, and finally the original vibration signal is decomposed into K IMF components; after performing a Hilbert transform on each modal component, the envelope spectrum is calculated, and the theoretical fault characteristic frequency and the amplitude of its harmonics are found and summed from the envelope spectra of each modal component to obtain the eigenvector. In the eigenvector, the denominator E is the amplitude of the main axis rotation frequency, and the numerator is the summation amplitude; this eigenvector serves as the observation input value when building the subsequent model; the parameter indices are in ascending order, and indices greater than 0.75 are selected as the initial reference indices.

5. The quality inspection method for rolling bearings according to claim 1 or 2, characterized in that, In step four, scrapped rolling bearings obtained during the maintenance process are used as samples to determine the effectiveness of parameter indicators. Scrapped bearings are divided into three types: bearings with obvious defects and scratches, bearings that have been scrapped due to their service life and have minor surface defects, and bearings that have been scrapped due to their service life but have good surface condition. Steps two and three are used to conduct quality inspections on the three types of scrapped bearings to obtain the pass rate of each type of scrapped bearing. The pass rate is compared with the set pass rate of each type of bearing, and then the magnitude of the parameter indicators is adjusted.

6. The quality inspection method for rolling bearings according to claim 1 or 2, characterized in that, Step five specifically involves: Step S5-1: Obtain the feature vector through step three. The Lloyd algorithm is used to scalar quantize the feature vectors to form the observation sequence. The initial probability π, the initial state transition matrix A, and the initial observation probability matrix B are obtained by a random function and used as the parameters of the HMM model. Step S5-2: Train the HMM model using the Baum-Welch algorithm; during the training process, the wear degree of each component of the bearing is divided into four states, namely normal state, inner ring damage state, outer ring damage state and rolling element damage, which represent the four hidden states of the HMM model, denoted as λ1, λ2, λ3 and λ4 respectively; the observation sequence of the HMM model is the observation sequence provided in step S5-1; Each set of feature vectors is vector quantized and used as the feature values ​​for training the HMM model using the Baum-Welch algorithm; after training, the HMM models corresponding to the four states are obtained. Step S5-3: When inspecting the quality of bearings produced in the same batch, sample bearings are selected, vibration signals are collected, feature vectors are obtained, and the feature vectors are vector-quantized. The vector-quantized feature vectors are then input into the HMM model for each state. Each HMM model outputs a log-likelihood probability lnp(O|λ), which represents the similarity between the observed sequence and the corresponding HMM model. The larger the value, the more similar the observed value is to the HMM model. The Viterbi algorithm is used to identify the optimal state sequence. The state corresponding to the largest log-likelihood probability is the damage state of the corresponding component of the bearing, thereby realizing the identification of the quality of each component of the bearing and determining the state of the bearing. Step S5-4: Through step S5-3, the damage level of the same batch of bearings is obtained as an important criterion for evaluating the entire production process of the production workshop, or the quality of bearings of the same type produced in different workshops is compared to select the superior workshop as a reference object.

Images