Bearing on-line health diagnosis system and bearing on-line health diagnosis method

Abstract

The invention provides a bearing on-line health diagnosis system and a bearing on-line health diagnosis method. A bearing on-line health diagnostic system includes a motor position encoder and a motor driver. The motor position encoder generates a position signal of the motor. The digital signal processing unit of the motor driver comprises a first digital differentiator and a second digital differentiator. The first digital differentiator performs a first digital differential process using the position signal to generate a speed signal. The second digital differentiator performs a second digital differential process on the speed signal to generate an acceleration signal. And the bearing on-line diagnosis function module receives the speed signal and the acceleration signal and takes the acceleration signal as a bearing signal to calculate the health state of the bearing.

Description

Technical Field

[0001] This case relates to a bearing diagnosis system and method, and more particularly to an online bearing health diagnosis system and an online bearing health diagnosis method. Background Art

[0002] In the prior art, it is necessary to install an additional sensor near the motor bearing, and the external sensor can only detect the signal of the motor bearing after capturing the sensing data. However, this method not only requires the installation of additional sensors to increase hardware costs, but also cannot detect the state of the bearing in real time due to the additional sensors and computing equipment.

[0003] Considering that the bearing is in the process of being damaged and the signal characteristics are usually not obvious and difficult to detect, how to efficiently and timely detect the health status of the bearing is the technical problem to be solved in this case. Summary of the invention

[0004] According to an embodiment of the present case, an online health diagnosis system for bearings is disclosed, which includes a motor position encoder and a motor driver. The motor driver includes a digital differentiator of a digital signal processing unit and a bearing online diagnosis function module. The bearing online diagnosis function module includes a bandpass signal filtering module, a feature conversion module, a healthy bearing model, an outlier calculation module, a signal envelope processing module, a frequency order analysis function module, and a bearing fault analysis module. The online health diagnosis system operates as follows: the position signal of the motor position encoder is transmitted to the digital signal processing unit on the motor driver via a transmission line and a communication protocol. The digital signal processing unit of the motor driver generates a speed signal through a first digital differential processing, and the speed signal generates an acceleration signal through a second digital differential processing. Both the speed signal and the acceleration signal enter the bearing online diagnosis function module, and the acceleration signal of the motor position encoder is used as a bearing signal. The bearing health status is output through the following bearing online health diagnosis method.

[0005] According to an embodiment of the present case, a method for online health diagnosis of a bearing is disclosed, comprising receiving a bearing signal and filtering the bearing signal with K filter frequency band parameters to obtain K filtered signals; performing N feature conversion calculations on each filtered signal through a feature conversion module to obtain K conversion signals; using a healthy bearing model, using machine learning to establish K model boundaries of a healthy bearing for M healthy data point groups in an N-dimensional space corresponding to M groups of healthy bearing signals; using an outlier calculation module, comparing L test data point groups in an N-dimensional space corresponding to L groups of test bearing signals with M data point groups at K model boundaries, respectively. To obtain K outliers, determine the maximum value from the K outliers, and select the filter frequency band parameter corresponding to the maximum value as the optimal filter parameter; through the signal envelope processing module, the continuous bearing signal to be tested that is continuously detected by the filter signal processing module is subjected to envelope operation on the continuous filter signal to be tested that is calculated through the optimal filter parameter to obtain an envelope spectrum signal; through the frequency conversion order analysis function module, the envelope spectrum signal is converted into an order spectrum analysis signal based on the bearing speed; and through the bearing fault analysis module, the order spectrum analysis signal is compared with the preset bearing fault signal to generate a bearing health diagnosis.

[0006] The second embodiment of this case discloses another method of the bearing online health diagnosis system: when the bearing to be diagnosed is not the bearing of the motor itself, but other bearings on the mechanism driven by the motor, because its transmission mechanism is connected to the motor, the rotation speed of the bearing to be diagnosed forms a fixed ratio with the motor speed, so the speed signal of the motor position encoder can be used to generate the speed signal of the bearing to be diagnosed through mechanical transmission ratio calculation. A vibration sensor is installed near the bearing to be diagnosed, and the bearing vibration signal is transmitted to the signal multiplexer of the motor driver. The vibration signal is used as the bearing signal, and both the bearing signal and the speed signal enter the bearing online diagnosis function module, and the bearing health status is output through the above-mentioned bearing online health diagnosis method calculation. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1A is a block diagram of an online bearing health diagnosis system according to an embodiment of the present invention;

[0008] Figure 1B is a block diagram of an online bearing health diagnosis system according to another embodiment of the present invention;

[0009] Figure 2 is a schematic diagram of a filter bank according to an embodiment of the present invention;

[0010] Figure 3 is a schematic diagram of a hypersphere according to an embodiment of the present invention;

[0011] Figure 4 (a) to Figure 4(d) is a schematic diagram of different outliers according to various embodiments of the present invention;

[0012] Figure 5 (a) to Figure 5 (c) is a schematic diagram of envelope calculation according to different embodiments of the present invention;

[0013] Figure 6 (a) Figure 5 (c) The two-dimensional signal diagram (frequency and amplitude) is combined with the bearing information (frequency and rotation speed (rpm)) to form a three-dimensional signal diagram;

[0014] Figure 6 (b) is a three-dimensional signal diagram of the envelope spectrum signal converted into an order spectrum analysis signal;

[0015] Figure 7 (a) shows an envelope spectrum signal corresponding to a center frequency of 10000 Hz and a bandwidth of 1000 Hz;

[0016] Figure 7 (b) shows the envelope spectrum signal corresponding to a center frequency of 21500 Hz and a bandwidth of 1000 Hz; Figure 8 is a flow chart of a bearing online health diagnosis method according to an embodiment of the present invention; Fig. 9 The figure is a flow chart of a bearing online health diagnosis method according to an embodiment of the present invention.

[0017]

Explanation of symbols

[0018] 210: Motor position encoder

[0019] 220: Motor driver

[0020] 230: Digital signal processing unit

[0021] 234: first digital differentiator

[0022] 236: Second digital differentiator

[0023] 240: Bearing online diagnosis function module

[0024] 110: Bandpass signal filtering module

[0025] 120: Feature conversion module

[0026] 130: Healthy Bearing Model

[0027] 140: Outlier calculation module

[0028] 150:Signal envelope processing module

[0029] 160: Frequency order analysis function module

[0030] 170: Bearing fault analysis module

[0031] 310: Hypersphere

[0032] 321, 323, 325: points

[0033] 411, 412, 413, 414: Health data points

[0034] 421, 422, 423, 424: data points to be tested

[0035] 431, 432, 433, 434: Bearing signals to be tested

[0036] 441, 442, 443, 444: Healthy bearing signals

[0037] 511, 512: Time segment

[0038] 513: Envelope spectrum signal

[0039] 514: Continuous filter signal to be tested

[0040] 521: characteristic frequency

[0041] 610: envelope spectrum signal

[0042] 620: Order spectrum analysis signal

[0043] 711, 721: Features

[0044] a: Center point

[0045] R: Hypersphere radius

[0046] S810~S850, S910~S940: Steps DETAILED DESCRIPTION

[0047] The present invention will be further described below in conjunction with the accompanying drawings and embodiments so that relevant persons in the technical field to which the present invention belongs can better understand the present invention and implement it accordingly, but the embodiments are not intended to limit the present invention.

[0048] Generally speaking, the health of a bearing will deteriorate as the operating time increases. The health of a bearing can be determined by detecting whether the bearing signal is abnormal. The wear state of a bearing is difficult to detect at the beginning of the product life cycle, which means that it is difficult to detect the signal characteristics of a suspected or impending failure in the bearing signal.

[0049] Furthermore, when the bearing is operating at a constant speed, the current health status of the bearing can be evaluated by referring to the normal bearing signal. However, when the bearing is operating at a variable speed (non-constant speed), if the normal bearing signal is directly used to evaluate the health status of the bearing, the diagnostic result will be inaccurate due to the difference in comparison basis.

[0050] The purpose of the bearing online health diagnosis technology in this case is to provide real-time and continuous detection of the bearing health status diagnosis, and is compatible with the health status diagnosis of bearings operating at non-constant speeds.

[0051] Figure 1A FIG. 4 is a block diagram of an online bearing health diagnosis system according to an embodiment of the present invention.

[0052] The bearing online health diagnosis system includes a motor position encoder 210 and a motor driver 220 . The motor driver 220 is connected to the motor position encoder 210 .

[0053] The motor driver 220 includes a digital signal processing unit 230 and a bearing online diagnosis function module 240 .

[0054] The digital signal processing unit 230 includes a first digital differentiator 234 and a second digital differentiator 236 .

[0055] The motor position encoder 210 is configured to generate a motor position signal and transmit the position signal to the digital signal processing unit 230 via a transmission line and a communication protocol.

[0056] The first digital differentiator 234 of the digital signal processing unit 230 is configured to perform a first digital differential process using the position signal to generate a speed signal. The first digital differentiator 234 transmits the speed signal to the second digital differentiator 236.

[0057] The second digital differentiator 236 is connected to the first digital differentiator 234. The second digital differentiator 236 is configured to perform a second digital differential process on the velocity signal to generate an acceleration signal.

[0058] The speed signal and the acceleration signal of the bearing are transmitted to the bearing online diagnosis function module 240. The bearing online diagnosis function module 240 is configured to receive the speed signal and the acceleration signal, and use the acceleration signal as a bearing signal to calculate the bearing health status.

[0059] The bearing online diagnosis function module 240 includes a bandpass signal filtering module 110 , a feature conversion module 120 , a healthy bearing model 130 , an outlier calculation module 140 , a signal envelope processing module 150 , a rotation frequency order analysis function module 160 and a bearing fault analysis module 170 .

[0060] The bandpass signal filtering module 110 is used to receive the bearing signal and perform filtering processing on the bearing signal using K filtering frequency band parameters to obtain K filtered signals, where K is a positive integer.

[0061] In one embodiment, the filter band parameter includes a filter band formed by a center frequency and a bandwidth. The bandpass signal filtering module 110 determines different combinations of multiple center frequencies and multiple bandwidths (i.e., different filter bands), and uses different filter bands to filter the bearing signal. In this way, the bandpass signal filtering module 110 generates filter signals corresponding to different filter band parameters. For example, the bandpass signal filtering module 110 determines K filter band parameters, and uses the K filter band parameters to filter the bearing signal to generate K filter signals corresponding to the K filter band parameters.

[0062] Since the filter frequency band is used for filtering the bearing signal, the selection of the filter frequency band will affect whether the fault characteristics of the bearing can be presented after filtering. In one embodiment, the bandpass signal filtering module 110 locates the center frequency by using a tree-like bisection method or a trisection method (see Figure 2 Explanation), and then take the center frequency as the center and take a bandwidth as the filter band. Multiple filter bands with all the center frequencies and bandwidths are combined to obtain multiple different filter signals, and then different filter signals are used for detection to find the filter band that can highlight the characteristics of bearing faults.

[0063] The feature conversion module 120 is connected to the bandpass signal filtering module 110. The feature conversion module 120 is used to obtain K conversion signals by subjecting each of the K filter signals to N feature conversion calculations. Each filter signal undergoes N feature conversion calculations, so each conversion signal has N features, where N is a positive integer. Since each conversion signal has N features, the conversion signal will form a data point in the N-dimensional space. Therefore, the K conversion signals form a data point group in the N-dimensional space, that is, this data point group includes K data points.

[0064] In one embodiment, the parameters of the N feature conversion calculations include variance, skewness, kurtosis, Gini coefficient, impulse factor, negative entropy, or autoencoder.

[0065] The N feature transformation calculations are shown in Table 1.

[0066] Table 1:

[0067]

[0068] The healthy bearing model 130 is connected to the feature conversion module 120. The healthy bearing model 130 receives the healthy bearing signal and establishes a model boundary through machine learning.

[0069] In one embodiment, the bandpass signal filtering module 110 receives M groups of healthy bearing signals, where M is a positive integer. The bandpass signal filtering module 110 uses K filter band parameters to filter each group of healthy bearing signals in the M groups to obtain K healthy filtered signals. The bandpass signal filtering module 110 obtains K healthy filtered signals corresponding to each group of healthy bearing signals, and outputs M groups of healthy filtered signals (each group has K healthy filtered signals) to the feature conversion module 120.

[0070] The feature conversion module 120 receives M groups of healthy filter signals. For each group of K healthy filter signals, the feature conversion module 120 performs N feature conversion calculations to obtain K healthy conversion signals. Each healthy conversion signal has N features, so each healthy conversion signal forms a data point group in an N-dimensional space. For the M groups of healthy filter signals, the feature conversion module 120 performs the same processing on all healthy filter signals to obtain M healthy data point groups, and each healthy data point group is located in the N-dimensional space.

[0071] The healthy bearing model 130 is connected to the feature conversion module 120. The healthy bearing model 130 is used to establish K model boundaries of a healthy bearing by machine learning for M healthy data point groups in an N-dimensional space corresponding to M groups of healthy bearing signals, wherein each model boundary corresponds to a filter band parameter.

[0072] In one embodiment, the healthy bearing model 130 executes a machine learning algorithm to perform model training on the healthy data point group, so that the M healthy data point groups form a cluster in the feature space to obtain the model boundary. The healthy bearing model 130 can detect whether the features of the test data fall inside or outside the model boundary, and determine whether the test data is an outlier.

[0073] Assume that there are L groups of bearing signals to be tested that need to be detected, where L is a positive integer (in fact, the bearing signals are continuously output and detected, and the L groups of signals are only for example). First, the bandpass signal filtering module 110 performs filtering processing of K filtering frequency bands on each bearing signal to be tested, and obtains K filtered signals to be tested for each group of bearing signals to be tested. In other words, the bandpass signal filtering module 110 will output L groups of filtered signals to be tested (each group has K filtered signals to be tested). Then, the feature conversion module 120 performs N feature conversion calculations on each group of K filtered signals to be tested to obtain K conversion signals to be tested. Since each conversion signal to be tested has N features, the K conversion signals to be tested constitute a data point group in an N-dimensional space. In this embodiment, there are L groups of bearing signals to be tested to be processed, so the feature conversion module 120 will output L groups of data points to be tested corresponding to the L groups of bearing signals to be tested, and each group of data points to be tested is a data point in an N-dimensional space.

[0074] Machine learning is, for example, Support Vector Data Description (SVDD), and the present invention does not limit the machine learning algorithm.

[0075] The outlier calculation module 140 is connected to the feature conversion module 120 and the healthy bearing model 130. The outlier calculation module 140 is used to detect the continuously generated bearing signal to be tested. The outlier calculation module 140 receives K model boundaries of healthy bearings and L data point groups to be tested. Each model boundary is regarded as a hypersphere (having a center point and a hypersphere radius). The outlier calculation module 140 calculates the distance between all data points of each data point group to be tested and the center point of each model boundary. If the distance is greater than the hypersphere radius, it means that this data point is an outlier. Conversely, if the distance is less than or equal to the hypersphere radius, it means that this data point is an inlier (details are in Figure 3 illustrate).

[0076] The outlier calculation module 140 calculates the ratio of outliers in the test data point group to all data points as the outlier rate. In other words, the outlier calculation module 140 takes the L test data point groups as a whole (i.e., representing the state of the test bearing) and performs tests through K model boundaries respectively.

[0077] In one embodiment, the outlier calculation module 140 is used to compare each of the L test data point groups in the N-dimensional space corresponding to the L groups of test bearing signals with the M data point groups of the K model boundaries to obtain K outliers corresponding to each of the K model boundaries. In this embodiment, the outlier calculation module 140 performs outlier calculations of the K model boundaries on the L test data point groups respectively, and obtains K outliers representing the L test data point groups.

[0078] According to the above description, the outlier calculation module 140 calculates K outliers representing the L groups of data points to be tested. Next, the outlier calculation module 140 determines the maximum value from the K outliers. As described above, each model boundary corresponds to a filter band parameter (including a center frequency and a bandwidth). The outlier calculation module 140 selects the filter band parameter corresponding to the model boundary of the maximum outlier value as the optimal filter parameter (including the optimal center frequency and the optimal bandwidth).

[0079] As described above, by finding the best filter parameters, the online bearing health diagnosis system uses the best filter parameters for subsequent signal processing without trying all frequencies one by one for filtering processing.

[0080] In one embodiment, based on the continuous operation of the bearing, the online bearing health diagnosis system receives the continuous bearing signal to be tested obtained by subsequent continuous detection. The bandpass signal filtering module 110 performs filtering processing with the optimal filtering parameters on the continuous bearing signal to be tested to obtain the continuous filtered signal to be tested.

[0081] The signal envelope processing module 150 is connected to the outlier calculation module 140. The signal envelope processing module 150 is used to perform envelope operation on the continuous filter signal to be tested to obtain an envelope spectrum signal (see Figure 5 illustrate).

[0082] It is worth mentioning that when the bearing rotates at a constant speed, the general fault detection method can determine that there is a problem with the bearing by analyzing the periodic characteristics obtained from the bearing signal. However, when the bearing rotates at an irregular speed, the original periodic characteristics are difficult to be judged, which greatly reduces the effectiveness of evaluating the health status of the bearing.

[0083] Based on this, in order to effectively judge the health status of the bearing, the online bearing health diagnosis system of this case processes the envelope spectrum signal before evaluating the bearing health status.

[0084] The frequency order analysis function module 160 is connected to the signal envelope processing module 150. The frequency order analysis function module 160 is used to convert the envelope spectrum signal into an order spectrum analysis signal based on the bearing speed.

[0085] In one embodiment, the frequency conversion order analysis function module 160 calculates the order spectrum analysis signal according to formula (1):

[0086]

[0087] Among them, order is the order spectrum analysis signal.

[0088] In formula (1), the purpose of the rotational frequency order analysis function module 160 dividing the envelope spectrum signal by the bearing speed is to align the time point of the envelope spectrum signal corresponding to the fault feature with the time point of the bearing speed. After alignment, that is, after calculating the order spectrum analysis signal, the amplitude of the envelope spectrum signal can be aligned to the order spectrum analysis signal, and the factors related to the speed are removed, so that a regular signal feature can be presented (for details, see Figure 6 illustrate).

[0089] The bearing fault analysis module 170 is connected to the rotation frequency order analysis function module 160. The bearing fault analysis module 170 is used to compare the order spectrum analysis signal with the preset bearing fault signal to generate a bearing health diagnosis.

[0090] The components of the bearing include an inner ring, an outer ring, a ball and a retainer. In one embodiment, the pattern of the preset bearing fault signal includes the pattern of the inner ring fault signal, the outer ring fault signal, the ball fault signal and the retainer fault signal of the bearing. If the order spectrum analysis signal compared by the bearing fault analysis module 170 matches the preset bearing fault signal, it means that the bearing has begun to fail. At this time, the bearing fault analysis module 170 knows that the bearing has begun to fail based on the matching preset bearing fault signal, and it can be known that the location where the bearing has begun to fail can be known by comparing the matching fault signal type.

[0091] Figure 1B FIG. 4 is a block diagram of an online bearing health diagnosis system according to another embodiment of the present invention.

[0092] When the bearing to be diagnosed is not the bearing of the motor itself but other bearings on the mechanism driven by the motor, due to the characteristics of the transmission mechanism connected to the motor, the rotation speed of the bearing to be diagnosed forms a fixed ratio with the motor speed. Therefore, the speed signal of the bearing to be diagnosed can be obtained by calculating the mechanical transmission ratio through the speed signal of the motor position encoder 210.

[0093] To achieve this, Figure 1B The bearing online health diagnosis system further includes a vibration sensor 260 and a multiplexer 270 .

[0094] The vibration sensor 260 is disposed on the bearing and is configured to detect the vibration of the bearing to generate a bearing vibration signal.

[0095] The multiplexer 270 is connected to the second digital differentiator 236 and the bearing online diagnosis function module 240 . The multiplexer 270 is configured to switch from receiving the acceleration signal to receiving the bearing vibration signal, so as to output the bearing vibration signal to the bearing online diagnosis function module 240 .

[0096] The bearing online diagnosis function module 240 is configured to receive the speed signal and the bearing vibration signal and use the bearing vibration signal as the bearing signal to calculate the bearing health status.

[0097] Please refer to Figure 2 . Figure 2 FIG. 4 is a schematic diagram of a filter bank according to an embodiment of the present invention.

[0098] like Figure 2 As shown, the horizontal axis is the frequency range, and the vertical axis is the level of the filter bank and the frequency division (Δf) k The bandpass signal filtering module 110 uses a filter grouper to divide the frequency into multiple different combinations, and then uses all the different combinations to filter the bearing signal one by one. For example, the frequency segmentation (Δf) of level 0 is k is 1, that is, the bandpass signal filtering module 110 uses the entire frequency band (center frequency is ) performs filtering on the bearing signal; the frequency division (Δf)k of level 1 is 1 / 2, that is, the frequency band is divided into 2 segments, and the bandpass signal filtering module 110 divides the entire frequency band into two segments, respectively with the center frequency as And the frequency range is The frequency band and the center frequency are And the frequency range is The bearing signal is filtered by the frequency band (there are two filtered signals at this time); the frequency segmentation (Δf) of level k k For 2 -k , that is, the frequency band is cut into 2 -k The bandpass signal filtering module 110 is 2 -k The center frequency and corresponding frequency band are used to filter the bearing signal one by one (in this case, there are 2 -k filtered signal).

[0099] In this embodiment, the bandpass signal filtering module 110 combines a plurality of center frequencies and bandwidths in a tree-like manner to obtain as many diverse filtering signals as possible for subsequent testing to determine the best filtering frequency band.

[0100] See also Figure 3 . Figure 3FIG. 4 is a schematic diagram of a hypersphere according to an embodiment of the present invention.

[0101] According to the above embodiment, the hypersphere 310 is the model boundary of the healthy bearing model 130 established by machine learning. The hypersphere 310 includes a center point a and a hypersphere radius R. The hypersphere 310 includes a healthy data point group (e.g., point 321). The outlier calculation module 140 calculates the distance between all data points of the data point group to be tested (e.g., points 323 and 325) and the center point a of the model boundary, and evaluates whether the distance is greater than the radius R of the hypersphere 310. For example, the distance between point 323 and the center point a is less than the radius R of the hypersphere 310 (i.e., point 323 falls inside the hypersphere 310), so point 323 is an inner group point. On the other hand, the distance between point 325 and the center point a is greater than the radius R of the hypersphere 310 (i.e., point 325 falls outside the hypersphere 310), so point 325 is an outlier.

[0102] It is worth noting that Figure 3 The hypersphere 310 corresponds to a filter frequency band parameter. Therefore, according to the above embodiment, the healthy bearing model 130 includes K model boundaries corresponding to a filter frequency band parameter, that is, there are K hyperspheres. Figure 3 Just one of them is shown for ease of understanding.

[0103] See also Figure 4 (a) to Figure 4 (d). Figure 4 (a) to Figure 4 (d) is a schematic diagram of different outliers according to multiple embodiments of the present invention.

[0104] According to the above embodiment, the outlier calculation module 140 calculates the ratio of outlier points in the data point group to be tested to all data points as outlier values.

[0105] like Figure 4 As shown in (a), the healthy bearing signal 441 is converted into the healthy data point 411 of the hyperplane. On the other hand, the bearing signal 431 to be tested is converted into the data point 421 to be tested. In this embodiment, all the data points to be tested fall within the hypersphere, so the outlier value is 0.

[0106] like Figure 4 As shown in (b), the healthy bearing signal 442 is converted into the healthy data point 412 of the hyperplane. On the other hand, the bearing signal 432 to be tested is converted into the data point 422 to be tested. In this embodiment, there is some noise in the bearing signal 432 to be tested, that is, some of the data points to be tested fall outside the hypersphere and some fall inside the hypersphere, so the outlier value is calculated to be 0.3.

[0107] like Figure 4As shown in (c), the healthy bearing signal 443 is converted into the healthy data point 413 of the hyperplane. On the other hand, the bearing signal 433 to be tested is converted into the data point 423 to be tested. In this embodiment, the bearing signal 433 to be tested has relatively regular characteristics, indicating that there may be a fault problem. At this time, all the data points 432 to be tested fall outside the hypersphere, so the outlier value is calculated to be 1.

[0108] like Figure 4 As shown in (d), the healthy bearing signal 444 is converted into the healthy data point 414 of the hyperplane. On the other hand, the bearing signal 434 to be tested is converted into the data point 424 to be tested. In this embodiment, there are some noises in the bearing signal 432 to be tested, which may be interference in the environment and belong to occasional changes. At this time, a few data points 441 to be tested fall outside the hypersphere and most of the data points 441 to be tested fall inside the hypersphere, so the outlier value is calculated to be 0.1.

[0109] It is worth mentioning that Figure 4 The outlier value of (c) is the maximum value, i.e. Figure 4 The model boundary of (c) can best present the fault signal. Therefore, the outlier calculation module 140 will correspond to Figure 4 The filter band parameters at the model boundary of (c) are taken as the optimal filter parameters.

[0110] See also Figure 5 (a) to Figure 5 (c). Figure 5 (a) to Figure 5 (c) is a schematic diagram of envelope calculation according to different embodiments of the present invention.

[0111] like Figure 5 As shown in (a), the signal envelope processing module 150 performs envelope operation on the continuous filter signal 514 to obtain an envelope spectrum signal 513. The time segment 511 is the inverse of the natural vibration frequency of the continuous filter signal 514, and the time segment 512 is the peak interval.

[0112] like Figure 5 As shown in (b), after the operation of the signal envelope processing module 150, an envelope spectrum signal 513 corresponding to the continuous tested filtered signal 514 is obtained.

[0113] like Figure 5 As shown in (c), the envelope processing module 150 converts the envelope spectrum signal 513 into a characteristic frequency 521 including multiple peaks based on the natural vibration frequency of the bearing.

[0114] The following is a schematic diagram of converting an envelope spectrum signal into an order spectrum analysis signal according to an embodiment of the present invention.

[0115] See also Figure 6 (a). Figure 6 (a) Figure 5 (c) The two-dimensional signal diagram (frequency and amplitude) is combined with the bearing information (frequency and rotation speed (rpm)) to form a three-dimensional signal diagram. Considering the maximum vibration frequency, the frequency of the envelope spectrum signal 610 increases with the increase of the rotation speed. Therefore, the three-dimensional signal diagram presents irregular changes.

[0116] See also Figure 6 (b). Figure 6 (b) is a three-dimensional signal diagram of the envelope spectrum signal 610 converted into an order spectrum analysis signal 620. In this embodiment, the rotational order analysis function module 160 calculates the order spectrum analysis signal 620 according to formula (1). Since the order spectrum analysis signal 620 removes the factors that depend on the rotation speed of the bearing, a signal characterized by regularity can be obtained. Considering the maximum vibration frequency, the order spectrum analysis signal 620 remains at the same order when the rotation speed increases. Accordingly, the three-dimensional signal diagram presents regular changes, and the fault characteristics of the bearing can be obtained through the regular changes.

[0117] Please refer to Figure 7 (a) and Figure 7 (b). Figure 7 (a) and Figure 7 (b) shows the effect of using different filtering frequencies in this case on determining whether a bearing fault occurs.

[0118] Figure 7 (a) shows the envelope spectrum signal corresponding to a center frequency of 10000 Hz and a bandwidth of 1000 Hz. The envelope spectrum signal in the frequency domain does not show an obvious characteristic frequency (such as feature 711). In contrast, Figure 7 (b) shows the envelope spectrum signal corresponding to a center frequency of 21500 Hz and a bandwidth of 1000 Hz. The envelope spectrum signal in the frequency domain shows an obvious characteristic frequency (such as characteristic 721). In other words, Figure 7 (b) The envelope spectrum signal is compared with Figure 7 The envelope spectrum signal of (a) can better highlight the fault characteristics of the bearing.

[0119] Please refer to Figure 8 , which is a flow chart of a bearing online health diagnosis method according to an embodiment of the present invention. The bearing online health diagnosis method can be executed by the bearing online health diagnosis system of FIG. 1 .

[0120] In step S810 , a bearing signal is received by the bandpass signal filtering module 110 and the bearing signal is filtered using K filtering frequency band parameters to obtain K filtered signals.

[0121] In step S820 , the feature conversion module 120 performs N feature conversion calculations on each of the K filtered signals to obtain K converted signals.

[0122] In step S830 , the healthy bearing model 130 is trained and M healthy data point groups in N-dimensional space corresponding to M groups of healthy bearing signals are used to establish K model boundaries of the healthy bearing through machine learning.

[0123] In step S840 , the outlier calculation module 140 obtains K outlier values ​​corresponding to each of the K model boundaries for each of the L test data point groups.

[0124] In step S850 , the outlier calculation module 140 determines a maximum value from the K outlier values, and selects a filtering band parameter corresponding to the maximum value as an optimal filtering parameter.

[0125] In this way, the bearing online health diagnosis method of this case improves the accuracy of the filtering process used in subsequent fault detection by finding the optimal filtering parameters.

[0126] Please refer to Fig. 9 , which is a flow chart of a bearing online health diagnosis method according to an embodiment of the present invention.

[0127] In step S910 , the bandpass signal filtering module 110 continuously receives the bearing signal to be tested, and performs filtering processing with optimal filtering parameters on the bearing signal to be tested to obtain the filtered signal to be tested.

[0128] In step S920 , the signal envelope processing module 150 performs envelope operation on the continuous tested filtered signal to obtain an envelope spectrum signal.

[0129] In step S930 , the envelope spectrum signal is converted into an order spectrum analysis signal based on the bearing speed by the rotational frequency order analysis function module 160 .

[0130] In step S940 , the bearing fault analysis module 170 compares the order spectrum analysis signal with a preset bearing fault signal to generate a bearing health diagnosis.

[0131] Figure 1A and Figure 1B The online bearing health diagnosis system is used to perform the above Figure 8 and Fig. 9 Online bearing health diagnosis method.

[0132] In summary, the bearing online health diagnosis system and bearing online health diagnosis method of this case can perform real-time online detection of the bearing health status. In addition, this case does not require the installation of additional equipment to detect the condition of the bearing online, eliminating installation problems and creating cost advantages.

[0133] The above description is only a specific example of this case, and does not limit the scope of the patent application of this case. Therefore, all equivalent changes made using the content of this case are also included in the scope of this case and are hereby stated.

Claims

1. A bearing online health diagnosis system, characterized in that: include: a motor position encoder configured to generate a position signal of a motor; A motor driver connected to the motor position encoder and comprising a digital signal processing unit, wherein the digital signal processing unit comprises: a first digital differentiator configured to perform a first digital difference process using the position signal to generate a velocity signal; a second digital differentiator connected to the first digital differentiator and configured to perform a second digital differential process on the velocity signal to generate an acceleration signal; and A bearing online diagnosis function module is configured to receive the velocity signal and the acceleration signal and use the acceleration signal as a bearing signal to calculate a bearing health status.

2. The bearing online health diagnosis system according to claim 1, characterized in that: Also includes: a vibration sensor disposed on a bearing and configured to detect vibration of the bearing to generate a bearing vibration signal; and a multiplexer connected to the second digital differentiator, the vibration sensor and the bearing online diagnosis function module and configured to switch from receiving the acceleration signal to receiving the bearing vibration signal to output the bearing vibration signal to the bearing online diagnosis function module; The bearing online diagnosis function module is configured to receive the speed signal and the bearing vibration signal and use the bearing vibration signal as the bearing signal to calculate the bearing health status.

3. The bearing online health diagnosis system according to claim 1, characterized in that: The bearing online diagnostic function module is configured to perform the following operations: A bearing signal is received through a bandpass signal filtering module and the bearing signal is filtered using K filter frequency band parameters to obtain K filtered signals; Performing N feature conversion calculations on each of the K filtered signals through a feature conversion module to obtain K conversion signals, wherein each of the K conversion signals has N features and constitutes a data point group in an N-dimensional space, and the data point group includes K data points; Using a healthy bearing model, the M healthy data point groups in the N-dimensional space corresponding to the M groups of healthy bearing signals are used for machine learning to establish K model boundaries of a healthy bearing, wherein the K model boundaries correspond one by one to the K filter band parameters; By using an outlier calculation module, the L test data point groups in the N-dimensional space corresponding to the L groups of test bearing signals are respectively compared with the M data point groups of the K model boundaries to obtain K outlier values ​​corresponding to each of the K model boundaries for each of the L test data point groups, a maximum value is determined from the K outliers, and the filter band parameter corresponding to the maximum value is selected as an optimal filter parameter; A signal envelope processing module is used to continuously detect a bearing signal to be tested that is continuously detected by the filter signal processing module, and an envelope operation is performed on a filter signal to be tested that is continuously detected and obtained by calculating the optimal filter parameters to obtain an envelope spectrum signal; Converting the envelope spectrum signal into a first-order spectrum analysis signal based on a bearing speed through a frequency order analysis function module; and A bearing health diagnosis is generated by comparing the order spectrum analysis signal with a preset bearing fault signal through a bearing fault analysis module; wherein K, N, M and L are positive integers respectively.

4. The bearing online health diagnosis system according to claim 3, characterized in that: Each of the K filter frequency band parameters includes K combinations formed by a center frequency and a frequency bandwidth, and the bandpass signal filtering module performs filtering processing on the bearing signal based on the center frequency and the frequency bandwidth of the K combinations respectively.

5. The bearing online health diagnosis system according to claim 3, characterized in that: The outlier calculation module calculates the ratio of the number of data points of the L data point groups to be tested outside each hypersphere of the K model boundaries to all data points as the K outliers.

6. The bearing online health diagnosis system according to claim 3, characterized in that: The state of the preset bearing fault signal includes the states of an inner ring fault signal, an outer ring fault signal, a ball fault signal and a retainer fault signal of a pre-known bearing. The bearing fault analysis module compares the state of the preset bearing fault signal with the state that matches the order spectrum analysis signal to determine the bearing fault location associated with the continuous tested filter signal.

7. A method for online bearing health diagnosis, characterized in that: include: A bearing signal is received through a bandpass signal filtering module and the bearing signal is filtered using K filter frequency band parameters to obtain K filtered signals; Performing N feature conversion calculations on each of the K filtered signals through a feature conversion module to obtain K conversion signals, wherein each of the K conversion signals has N features and constitutes a data point group in an N-dimensional space, and the data point group includes K data points; Using a healthy bearing model, the M healthy data point groups in the N-dimensional space corresponding to the M groups of healthy bearing signals are used for machine learning to establish K model boundaries of a healthy bearing, wherein the K model boundaries correspond one by one to the K filter band parameters; By using an outlier calculation module, the L test data point groups in the N-dimensional space corresponding to the L groups of test bearing signals are respectively compared with the M data point groups of the K model boundaries to obtain K outlier values ​​corresponding to each of the K model boundaries for each of the L test data point groups, a maximum value is determined from the K outliers, and the filter band parameter corresponding to the maximum value is selected as an optimal filter parameter; A signal envelope processing module is used to continuously detect a bearing signal to be tested that is continuously detected by the filter signal processing module, and an envelope operation is performed on a filter signal to be tested that is continuously detected and obtained by calculating the optimal filter parameters to obtain an envelope spectrum signal; Converting the envelope spectrum signal into a first-order spectrum analysis signal based on a bearing speed through a frequency order analysis function module; and A bearing health diagnosis is generated by comparing the order spectrum analysis signal with a preset bearing fault signal through a bearing fault analysis module; wherein K, N, M and L are positive integers respectively.

8. The method for online health diagnosis of a bearing according to claim 7, characterized in that: It also includes K combinations formed by each of the K filter band parameters including a center frequency and a bandwidth, and the bandpass signal filtering module performs filtering processing on the bearing signal based on the center frequency and the bandwidth of the K combinations respectively.

9. The method for online health diagnosis of a bearing according to claim 7, characterized in that: The method also includes calculating, by the outlier calculation module, the ratio of the number of data points of the L test data point groups outside each of the K model boundary hyperspheres to all data points as the K outliers.

10. The method for online health diagnosis of a bearing according to claim 7, characterized in that: The state of the preset bearing fault signal includes the states of an inner ring fault signal, an outer ring fault signal, a ball fault signal and a retainer fault signal of a pre-known bearing. The online bearing health diagnosis method also includes comparing the state of the preset bearing fault signal with the state that matches the order spectrum analysis signal through the bearing fault analysis module to determine the bearing fault location associated with the continuous tested filter signal.

Images