JP7716309B2 - Wheel section abnormality detection method and abnormality detection device - Google Patents

Description

This invention relates to an abnormality detection technology for detecting abnormalities such as contact between parts in the wheel section of a vehicle such as an automobile based on the sound generated in the wheel section.

For example, abnormal noise can occur when a back plate, which is installed to protect the disc rotor of a brake device, comes into contact with the disc rotor. It is generally difficult for workers to accurately determine abnormalities in the contact of parts with rotating objects.

Patent Document 1 discloses a technology for detecting abnormal vibrations in wheel bearings and preventing serious accidents by installing acceleration sensors on the bearings or their surrounding components, as well as installing an acceleration sensor at a reference location that is located away from the bearing and is not affected by the vibrations of the bearings. The signals from each sensor are subjected to FFT processing under specific driving conditions (for example, driving straight at 40 km/h), and the presence or absence of an abnormality is determined by comparing the peak levels of the frequency spectrum between the two.

JP 2016-38269 A

The above technology requires acceleration sensors at the vibrating part and the reference part, and if there are multiple parts that could be the cause of vibration, acceleration sensors must be installed on all of them. Therefore, in reality, it cannot be applied to inspections where a single target part is not identified. Furthermore, if the candidate parts are close to each other, it becomes difficult to separate the vibrations from each other.

The method for detecting an abnormality in a wheel portion according to the present invention comprises:
Acquire sounds generated near the wheels of a target vehicle while changing the vehicle speed to generate sound data;
This sound data is converted according to frequency to generate frequency data associated with vehicle speed ,
This frequency data is compared with basic frequency data associated with the corresponding vehicle speed to extract abnormal noise.
The frequency characteristics of the abnormal noise are compared with the natural frequency of the part to identify the candidate part, and the correlation between the sound characteristics of the abnormal noise and the vehicle speed is used to determine whether the part in question is abnormal.
The results including the identified parts are presented.

According to this invention, by focusing on the correlation between the sound characteristics of the abnormal sound and the vehicle speed , it is possible to reliably identify whether the abnormal sound is an abnormal sound or a disturbance noise, and by comparing the frequency characteristics of the abnormal sound with the natural frequency of the part, it is possible to identify the part in an abnormal state. Therefore, it is possible to detect abnormalities in multiple parts simply by acquiring the sound.

1 is a functional block diagram of a first embodiment in which the present invention is applied to a finished vehicle inspection process. 4 is a flowchart showing the flow of processing in the first embodiment. FIG. 10 is an explanatory diagram of an example of a spectrogram displayed as frequency data. FIG. 10 is a functional block diagram of a second embodiment. 10 is a flowchart showing the flow of processing according to a second embodiment. FIG. 10 is a functional block diagram of a third embodiment. 10 is a flowchart showing the flow of processing according to a third embodiment. FIG. 10 is a functional block diagram of a fourth embodiment. 10 is a flowchart showing the flow of processing according to a fourth embodiment. FIG. 13 is a functional block diagram of a fifth embodiment. 13 is a flowchart showing the flow of processing according to the fifth embodiment. 13 is a flowchart showing another example of the processing flow of the fifth embodiment. 1 is a characteristic diagram showing examples of linear correlation and nonlinear correlation. FIG. 4 is an explanatory diagram showing a display example on a display unit.

One embodiment of the present invention will now be described. This embodiment of wheel section abnormality detection is performed, for example, during the finished vehicle inspection process, the final stage of an automobile production line, as an inspection to confirm that there is no contact or interference between parts and rotating bodies in the wheel section. Generally, during the finished vehicle inspection process, an inspector test-drives the finished vehicle to be inspected on free rollers, inspecting numerous items such as the engine, meters, and brakes. During this test run on the free rollers, the vehicle is accelerated for an appropriate inspection time (e.g., approximately 15-20 seconds) until it reaches an appropriate high speed (e.g., 120 km/h), and an abnormality inspection is performed based on the sounds generated from the wheel section during this time. Hereinafter, the abnormality detection device of this embodiment will also be referred to as a "wheel contact inspection device," and this wheel contact inspection device is configured as part of the inspection equipment in the finished vehicle inspection process.

Figure 1 shows a functional block diagram of the wheel contact inspection device of the first embodiment. The wheel contact inspection device of the first embodiment is composed of a sound measurement/acquisition unit 10, a vehicle motion state acquisition unit 70, a frequency conversion unit 20, an abnormal sound determination unit 30, a natural frequency acquisition unit 40, an abnormal sound component identification unit 50, and a display unit 60.

The sound measurement and acquisition unit 10 includes a microphone that acquires sound generated from the wheel section and converts it into an electrical signal, i.e., sound data, and a recording unit that temporarily stores this sound data. The microphone is placed outside the vehicle so that it can collect sound from the wheel section of the vehicle as it runs on free rollers. The microphone's directivity and frequency characteristics are selected according to its position relative to the vehicle being measured and the required frequency band. Typically, a microphone with directivity toward the wheel section of the vehicle is used. A microphone array or the like can be used to localize the sound source and obtain sound data from which noise has been removed. It is also possible to use an existing microphone, such as one installed to collect horn sounds during the finished vehicle inspection process, for this microphone.

As mentioned above, the wheel contact inspection is performed while increasing the vehicle speed for, for example, 15 to 20 seconds, so sound data with a duration of, for example, 15 to 20 seconds is acquired. Note that because the vehicle speed changes over time, the sound data will include sounds made at various different vehicle speeds. Furthermore, the sound data acquired during the test run will also include engine sounds, etc.

The vehicle motion state acquisition unit 70 acquires the vehicle motion state, in this embodiment, the vehicle speed, from the signal of the vehicle speed sensor on the vehicle side, the rotation of the free rollers, etc., and outputs it to be associated with the sound data .

The frequency conversion unit 20 generates frequency data by converting the acquired sound data according to frequency using frequency analysis techniques such as FFT (Fast Fourier Transform) and wavelet analysis. In particular, since it is necessary to ultimately determine the time change in sound pressure of a specific sound (change with respect to vehicle speed), frequency analysis is performed for each relatively short time domain to generate frequency data for each individual time.

When frequency data is expressed in two dimensions, for example, it is expressed as a frequency spectrum with frequency on the horizontal axis and sound pressure (or power) on the vertical axis. Furthermore, by overlaying the conversion results for each time in a time series, the frequency data can be handled as a so-called spectrogram, which is three-dimensional data that includes time.

Figure 3 shows an example of a spectrogram display created by processing sound data obtained from a wheel contact inspection using FFT. In accordance with the general spectrogram format, the horizontal axis represents time and the vertical axis represents frequency. The total time length is approximately 15 seconds. Frequencies up to approximately 10 kHz are included. The brightness and color of each point represent the power (dBA) (or sound pressure or amplitude) of a given frequency at a given time. While the attached image is black and white, an actual spectrogram display uses different colors, such as red for high power, yellow-green for medium power, and blue for low power, and the power is indicated by varying brightness within each color. The clustered sound at the bottom of the spectrogram in Figure 3 (designated E) represents the engine sound during acceleration. The engine sound suddenly decreases at time t1, as acceleration ends.

Note that the graph shown below the spectrogram display in Figure 3 represents the original sound data, with amplitude on the vertical axis.

The abnormal sound detection unit 30 extracts abnormal sounds by comparing frequency data associated with vehicle speed with basic frequency data. The abnormal sound detection unit 30 includes a memory unit, which pre-stores frequency data for normal vehicles at each vehicle speed as basic frequency data. In one example, the frequency data is a frequency spectrum of sound pressure vs. frequency, and the data of the vehicle being inspected is compared with the basic data to extract differences in sound pressure peaks as abnormal sounds. In other words, sound pressure peaks not included in the basic frequency data are abnormal sounds. The influence of the engine sound during acceleration described above can basically be eliminated by comparing it with this basic frequency data associated with vehicle speed. Note that the three-dimensional spectrogram described above may be used instead of a two-dimensional frequency spectrum, and abnormal sounds may be extracted by comparing it with the basic spectrogram.

The natural frequency acquisition unit 40 contains a parts database that compiles the names, part numbers, and natural frequencies of the numerous parts in the wheel section that are the target of abnormality detection, and outputs this information to the abnormal noise part identification unit 50. The natural frequency of each part is measured using a hammering test or other method on the actual target part, or a value theoretically calculated using CAE. The natural frequency is a value specific to the part, and is uniquely determined by the part's material, shape, fixing method, etc.

The abnormal noise component identification unit 50 compares the frequency characteristics of the extracted abnormal noise with the natural frequencies of multiple components included in the wheel assembly to tentatively identify candidate components, and determines whether the component is abnormal based on the correlation between the sound characteristics of the abnormal noise and the vehicle's motion state. Essentially, components with a natural frequency (or its overtones) roughly equal to the frequency of the sound pressure peak of the abnormal noise can be candidates for the source of the abnormal noise. If the abnormal noise is actually caused by vibration of this candidate component (for example, vibration due to contact with a rotating body such as a disc rotor), then the sound characteristics (e.g., sound pressure) of the abnormal noise should change when the motion state of the vehicle, which is the source of the vibration, changes. Therefore, the abnormal noise component identification unit 50 calculates the correlation between the sound pressure (peak sound pressure) of the abnormal noise corresponding to the natural frequency of the component selected as a candidate and vehicle speed, for example by linear regression, and determines whether there is a correlation between the two based on the regression coefficient.

If the noise extracted as an abnormal sound does not correlate with the vehicle's motion state, such as vehicle speed, the sound is likely to be disturbance noise and is not treated as an abnormal part. If there is a certain correlation between the two, the part selected as a candidate is ultimately determined to be abnormal (in this example, a part in contact with a rotating body).

In the spectrogram shown in Figure 3, an example, a continuous abnormal noise with a constant frequency labeled "BP" can be seen. Although it is not clear in the black and white image of Figure 3, the line representing this abnormal noise changes color to indicate an increase in sound pressure as the vehicle speed increases (as time passes up to t1). Furthermore, because the tires continue to rotate after time t1, it is shown that the abnormal noise continues, in contrast to the sudden decrease in engine sound (E) at time t1.

The display unit 60 is a display means for displaying the inspection results to relevant parties such as workers near the inspection device, inspectors driving the vehicles, managers, and data scientists who will use the data. It may include, for example, an LCD display or an organic EL display. Furthermore, if any alarm sounds or audio are involved, it may also include a sound source, amplifier, speaker, etc. for generating and emitting the sound. The images, videos, sounds, and audio displayed here are generated by image generation means and sound/audio generation means included in the display unit 60. Specific examples of what is displayed on the display will be explained later.

Figure 2 is a flowchart showing the processing flow of the wheel contact inspection device of the first embodiment described above. First, the microphone of the sound measurement and acquisition unit 10 collects sound near the wheel section of a vehicle test-running on free rollers and acquires it as sound data (Step 1). Next, the frequency conversion unit 20 described above converts the sound data according to frequency using frequency analysis techniques such as FFT (Fast Fourier Transform) and wavelet analysis to generate frequency data. More specifically, frequency analysis is performed for each relatively short time period to generate frequency data for each time period (Step 2). Additionally, information on the vehicle speed at the time the wheel section sound is acquired is read as the vehicle motion state (Step 3).

Next, the frequency data associated with the vehicle speed (e.g., a two-dimensional frequency spectrum) is compared with the basic frequency data corresponding to the same vehicle speed, and the difference between the two, i.e., the presence or absence of sound pressure peaks that do not exist in the basic frequency spectrum, is determined (step 4). If there is no difference between the two, the system proceeds to step 10, determines that the data is normal, and displays a message to that effect.

If there is a discrepancy between the measured and the base frequency data, the system retrieves information on the natural frequencies of multiple components near the wheel assembly from the components database (Step 5), then proceeds to Step 6 to determine whether the noise originates from a component in the wheel assembly and identify the component that is producing the abnormal noise. Specifically, the system compares the frequency of the sound pressure peak that is the abnormal noise with the natural frequencies in the components database to identify multiple candidate components. Using multiple frequency data associated with vehicle speed, the system then calculates the correlation between vehicle speed and sound pressure for the natural frequencies of the candidate components (including harmonics) using linear regression, determining the regression coefficient α1 (when the regression line is y = α1x + β1) (Step 6). The existence of a correlation between the two is then determined based on whether this regression coefficient α1 is equal to or greater than a predetermined threshold (Step 7). In other words, if the abnormal noise is due to contact between components, such as between a backplate and a disc rotor, the system focuses on the relationship that as vehicle speed increases, the energy applied to the components increases and the sound pressure increases. If the sound pressure does not change even with increasing vehicle speed, the system determines that the noise is due to some kind of external noise. This linear regression calculation and the magnitude judgment of the regression coefficient α1 are performed for each of the multiple parts. If there are no parts for which the regression coefficient α1 is equal to or greater than the threshold, the system proceeds to step 10, determines that the part is normal, and displays a message to that effect.

Here, if the regression coefficient α1 for a certain part (e.g., part A) that has a natural frequency corresponding to the abnormal noise is equal to or greater than the threshold value, the system proceeds to step 8, determines that part A is abnormal (i.e., the source of the abnormal noise), and displays a message to that effect.

Finally, in step 9, it is determined whether retesting is necessary for some reason. If retesting is necessary, return to step 1; if not, end the test.

As described above, the wheel contact inspection device of the above embodiment uses a microphone to collect sounds near the wheel section and processes the signals to detect contact between parts such as the back plate and rotating bodies such as the disc rotor. In particular, by utilizing information on the natural frequencies of multiple parts near the wheel section and the correlation between the sound pressure of the extracted abnormal noise and vehicle speed, it is possible to identify the part that is producing the abnormal noise due to contact while eliminating the influence of other noise, even in an environment where other noise is present. Therefore, wheel contact inspection can be performed efficiently during test runs on free rollers during the completed vehicle inspection process.

Next, a second embodiment of the wheel contact inspection device will be described with reference to Figures 4 and 5. The following mainly focuses on the differences from the first embodiment. The wheel contact inspection device of the second embodiment uses the half-width of the waveform in the frequency spectrum as the sound characteristic instead of sound pressure.

As shown in the block diagram of Figure 4, the wheel contact inspection device of the second embodiment, like the first embodiment, is configured to include a sound measurement/acquisition unit 10, a vehicle motion state acquisition unit 70, a frequency conversion unit 20, an abnormal sound determination unit 30, a natural frequency acquisition unit 40, an abnormal sound component identification unit 50, and a display unit 60.

Here, when detecting abnormal sounds, the abnormal sound detector 30 compares full widths at half maximum (FWHM) instead of comparing sound pressures as in the first embodiment. When a peak is detected in a specific frequency range in the sound pressure-frequency characteristics, the full width at half maximum is calculated by extracting the maximum (peak value) and minimum sound pressure values within that range and calculating the intermediate value H between the two (H = (maximum value - minimum value)/2). This full width at half maximum basically corresponds to the level of sound pressure at each frequency, and while the peak value of sound pressure varies depending on the shape of the peak, this full width at half maximum can represent the magnitude of energy at each frequency without being affected by the peak shape. Therefore, while the frequency data in the first embodiment is a frequency spectrum of sound pressure vs. frequency, the second embodiment uses a frequency spectrum of full width at half maximum vs. frequency, where the vertical axis is the full width at half maximum and the horizontal axis is the frequency.

In other words, a basic half-width-frequency spectrum associated with vehicle speed is stored in advance, and a half-width-frequency spectrum associated with vehicle speed is generated from the sound data of the vehicle being inspected. By comparing the two, peak components of the half-width that do not exist in the basic frequency data are extracted as abnormal noise.

Furthermore, instead of evaluating the correlation between sound pressure and vehicle speed as in the first embodiment described above, the abnormal noise component identification unit 50 evaluates the correlation between the half-width of the half-width peak frequency extracted as abnormal noise and vehicle speed. In other words, the correlation between the half-width of the abnormal noise corresponding to the natural frequency of the component selected as a candidate and vehicle speed is obtained, for example, by linear regression, and the existence or absence of a correlation between the two is determined from the regression coefficient.

Figure 5 is a flowchart showing the processing flow for wheel contact inspection in the second embodiment. After steps 1 to 3, where sound data is acquired, frequency data is generated, and vehicle speed is read, step 4A compares the half-width-frequency spectrum associated with vehicle speed with the basic half-width-frequency spectrum also associated with vehicle speed to extract abnormal noise. If there is no difference between the two, the system proceeds to step 10, determines that the system is normal, and displays a message to that effect.

If an abnormal noise is detected, the system acquires the component's natural frequency information in step 5, then proceeds to step 6A. The system compares the half-width peak frequency of the abnormal noise with the natural frequencies in the component database to identify multiple candidate components. Using a large amount of frequency data associated with vehicle speed, the system calculates the correlation between the half-width of the natural frequency of the candidate components (including harmonics) and vehicle speed using linear regression, and determines the regression coefficient α2 (when the regression line is y = α2x + β2) (step 6A). The system then determines whether the regression coefficient α2 is greater than or equal to a predetermined threshold value to determine whether there is a correlation between the two (step 7A). If no component has a regression coefficient α2 greater than or equal to the threshold value, the system proceeds to step 10, where the component is determined to be normal, and a message to that effect is displayed. If the regression coefficient α2 for a component (e.g., component A) with a natural frequency corresponding to the abnormal noise is greater than or equal to the threshold value, the system proceeds to step 8, where component A is determined to be abnormal (i.e., the source of the abnormal noise), and a message to that effect is displayed.

In this way, by using the half-width instead of sound pressure for the frequency data after frequency analysis, it becomes less affected by the waveform of the sound pressure peak, as mentioned above.

Next, a third embodiment of the wheel contact inspection device will be described with reference to Figures 6 and 7. The third embodiment of the wheel contact inspection device uses both sound pressure, as in the first embodiment, and the half-width of the frequency waveform, as in the second embodiment, as sound characteristics.

As shown in the block diagram of Figure 6, the wheel contact inspection device of the third embodiment, like the first embodiment, is configured to include a sound measurement/acquisition unit 10, a vehicle motion state acquisition unit 70, a frequency conversion unit 20, an abnormal sound determination unit 30, a natural frequency acquisition unit 40, an abnormal sound component identification unit 50, and a display unit 60.

When extracting abnormal sounds, the abnormal sound detector 30 performs both a comparison with the basic characteristics of sound pressure as in the first embodiment, and a comparison with the basic characteristics of full width at half maximum (FWHM) as in the second embodiment.

The abnormal noise component identification unit 50 then evaluates both the correlation between sound pressure and vehicle speed in the first embodiment described above, and the correlation between half-value width and vehicle speed in the second embodiment.

Figure 7 is a flowchart showing the processing flow for wheel contact inspection in the third embodiment. Steps 1 to 3 involve acquiring sound data, generating frequency data, and reading vehicle speed, and then step 5 involves acquiring information on the natural frequencies of parts around the wheel.

Next, in step 6, a large number of frequency data items associated with vehicle speed are used to calculate the correlation between sound pressure and vehicle speed for the natural frequency sounds (including harmonics) of a large number of components using linear regression, and the regression coefficient α1 (when the regression line is y = α1x + β1) is determined. In step 4, the sound pressure-frequency spectrum associated with vehicle speed is compared with the basic sound pressure-frequency spectrum corresponding to the same vehicle speed to determine whether there is a difference between the two. If there is a difference between the two, i.e., an abnormal sound, the presence or absence of a correlation between sound pressure and vehicle speed is determined for candidate components having a natural frequency corresponding to the frequency of the abnormal sound by determining whether the regression coefficient α1 is equal to or greater than a predetermined threshold (step 7).

If these determinations are YES, in step 6A, a large number of frequency data items associated with vehicle speed are used to calculate the correlation between vehicle speed and half-width for the sounds (including harmonics) of the natural frequencies of the large number of components using linear regression, and the regression coefficient α2 (when the regression line is y = α2x + β2) is determined. In addition, in step 4A, the frequency spectrum of half-width - frequency associated with vehicle speed is compared with the frequency spectrum of the basic half-width - frequency corresponding to the same vehicle speed to determine whether there is a difference between the two. If there is a difference between the two, i.e., an abnormal sound, the presence or absence of a correlation between half-width and vehicle speed is determined for candidate components having natural frequencies corresponding to this abnormal sound by determining whether the regression coefficient α2 is equal to or greater than a predetermined threshold (step 7A).

If the determinations in steps 4, 7, 4A, and 7A for a certain part A are all YES, then the part is determined to be abnormal (i.e., the source of an abnormal sound) and a message to that effect is displayed (step 8). If any of the results are NO, proceed to step 10, where the part is determined to be normal and a message to that effect is displayed.

In this third embodiment, the vehicle speed change in the sound pressure of the abnormal noise and the vehicle speed change in the half-width are used as an AND condition, making abnormality determination more strict and reducing false detections due to noise, etc.

In the processing flow shown in the flowchart of Figure 7, the regression coefficients α1 and α2 are calculated for all parts around the wheel before extracting abnormal noises by comparing them with the basic characteristics. As in the first and second embodiments described above, after extracting abnormal noises, the regression coefficients α1 and α2 may be calculated only for tentative candidate parts that have natural frequencies corresponding to the abnormal noises.

Also, in the above example, processing for sound pressure is performed first, but processing for half-width may also be performed first.

Next, a fourth embodiment of the wheel contact inspection device will be described with reference to Figures 8 and 9. The fourth embodiment of the wheel contact inspection device is designed to more accurately determine whether an abnormal noise originates from the wheel section by checking whether the abnormal noise contains a fluctuating sound component (a sound in which the sound of a predetermined high-frequency component fluctuates by approximately 1 to 10 Hz) that corresponds to the tire rotation speed.

For example, if the cause of the abnormal noise is interference between the back plate and the disc rotor, the abnormal noise occurs when part of the back plate comes into contact with the disc rotor with each rotation of the disc rotor (i.e., with each rotation of the tire), and in many cases the components of the abnormal noise include a fluctuating sound component corresponding to the rotational speed of the tire. For example, if the tire diameter is 64 cm and the vehicle speed is 10 km/h, the rotational frequency of the tire is approximately 1.3 Hz, so the vibration noise of a part with a natural frequency of, say, 5 kHz will include a fluctuating sound component of approximately 1.3 Hz.

As shown in the block diagram of Figure 8, the wheel contact inspection device of the fourth embodiment, like the first embodiment, is configured to include a sound measurement/acquisition unit 10, a vehicle motion state acquisition unit 70, a frequency conversion unit 20, an abnormal sound determination unit 30, a natural frequency acquisition unit 40, an abnormal sound component identification unit 50, and a display unit 60. It also includes a time-varying frequency calculation unit 80, which determines the fluctuating sound components of the acquired sound data and calculates whether a fluctuating frequency equivalent to the wheel rotation frequency is included. If a fluctuating component equivalent to the wheel rotation frequency is not included, it is determined that there is no contact with a rotating body at the wheel. The fluctuating component equivalent to the wheel rotation frequency is, for example, approximately 1 to 10 Hz, but this may be set to a narrower range taking into account the vehicle motion state, i.e., vehicle speed, or may be set to a fixed range (e.g., 1 to 10 Hz) regardless of vehicle speed.

Figure 9 is a flowchart showing the processing flow for wheel contact inspection in the fourth embodiment. Steps 1 to 3 involve acquiring sound data, generating frequency data, and reading vehicle speed, and then step 5 involves acquiring information on the natural frequencies of parts around the wheel.

Next, in step 11, the fluctuating sound component contained in the sound data, i.e., the time-varying frequency, is calculated based on the frequency data, and in step 12, it is determined whether this time-varying frequency is substantially the same as the rotation frequency of the wheel.

The time-varying frequency can be calculated using a known fluctuating sound analysis algorithm. For example, fluctuating sound can be analyzed by setting analysis frames in minute time units and using multiple band-pass filters for extracting fluctuating components to find the peaks and valleys of the waveform after passing through the band-pass filters for each analysis frame.

If it is determined in step 12 that the time-varying frequency is different from the wheel rotation frequency, proceed to step 10, determine that the condition is normal, and display a message to that effect. If the time-varying frequency is substantially the same as the wheel rotation frequency, proceed to step 6 and beyond.

The rest of the process is the same as in the previous embodiments. In step 6, a large number of frequency data items associated with vehicle speed are used to calculate the correlation between sound pressure and vehicle speed for the natural frequency sounds (including harmonics) of a large number of components using linear regression, and the regression coefficient α1 (when the regression line is y = α1x + β1) is determined. In step 4, the sound pressure-frequency spectrum associated with vehicle speed is compared with the basic sound pressure-frequency spectrum corresponding to the same vehicle speed to determine whether there is a difference between the two. If there is a difference between the two, i.e., an abnormal sound, the presence or absence of a correlation between sound pressure and vehicle speed is determined for candidate components having natural frequencies corresponding to the frequency of the abnormal sound by determining whether the regression coefficient α1 is equal to or greater than a predetermined threshold (step 7).

If the determinations in steps 4 and 7 for a certain part A are both YES, then the part is determined to be abnormal (i.e., the source of an abnormal sound) and a message to that effect is displayed (step 8). If either result is NO, proceed to step 10, where the part is determined to be normal and a message to that effect is displayed.

In the processing flow shown in the flowchart in Figure 9, the regression coefficient α1 is calculated for all parts around the wheel before extracting abnormal noises by comparing them with the basic characteristics. As in the first and second embodiments described above, after extracting abnormal noises, the regression coefficient α1 may be calculated only for tentative candidate parts that have natural frequencies corresponding to the abnormal noises.

Also, in the example flowchart above, only sound pressure is considered, but both sound pressure and half-width may be used, as in the third embodiment.

In this fourth embodiment, the presence of fluctuating sound components corresponding to the wheel rotation frequency is a weighting condition, making abnormality determination stricter than in the first and second embodiments, and reducing false detections due to noise, etc.

Next, a fifth embodiment of the wheel contact inspection device will be described with reference to Figures 10 to 13. The fifth embodiment of the wheel contact inspection device determines whether the correlation between the sound pressure of the natural frequency sound of a part and the vehicle motion state (e.g., vehicle speed) is simply linear or nonlinear (e.g., a quadratic function), allowing for more detailed abnormality inspection.

As mentioned above, when abnormal noise is caused by contact between a rotating body and a part, for example, as vehicle speed increases, the vibration energy acting on the part increases, and so sound pressure or half-width tends to increase with vehicle speed. However, the correlation between the two may not necessarily be linear depending on the material of the part, etc. Figure 13 shows the relationship between sound pressure or half-width and vehicle speed, with the characteristics of one part shown by line L1 being linear. On the other hand, the characteristics of another part shown by line L2 are nonlinear. More specifically, in the process of calculating the correlation coefficient, regression equations are calculated using both linear and nonlinear approximations, and whether the relationship is linear or nonlinear is determined by comparing the magnitude of each correlation coefficient. In the example of Figure 13, for example, the correlation coefficient of line L1 is "0.9714" and the correlation coefficient of line L2 is "0.9978." A comparison of the magnitude of the two indicates that the relationship is nonlinear. Note that a nonlinear relationship with vehicle speed may also be linear with other variables, such as vehicle acceleration.

As shown in the block diagram of Figure 10, the wheel contact inspection device of the fifth embodiment, like the first embodiment, is configured to include a sound measurement/acquisition unit 10, a vehicle motion state acquisition unit 70, a frequency conversion unit 20, an abnormal sound determination unit 30, a natural frequency acquisition unit 40, an abnormal sound component identification unit 50, and a display unit 60. It also includes a correlation order calculation unit 90 that determines whether the characteristics of the frequency components extracted as abnormal sounds relative to vehicle speed are linear or nonlinear. This correlation order calculation unit 90 calculates an approximate correlation curve and determines whether the characteristic is linear or nonlinear from the order. In this embodiment, whether the characteristics of a component are linear or nonlinear is determined from the materials contained in the component database of the natural frequency acquisition unit 40; however, the component database may also include information on whether each component is linear or nonlinear.

Figure 11 is a flowchart showing the processing flow for wheel contact inspection in the fifth embodiment. Steps 1 to 3 involve acquiring sound data, generating frequency data, and reading vehicle speed, and then step 5 involves acquiring information on the natural frequencies of parts around the wheel.

Next, in step 21, regression equations are calculated using both linear and non-linear approximations (quadratic functions) based on the frequency data corresponding to vehicle speed, and the respective correlation coefficients are calculated. Then, in step 22, the two correlation coefficients are compared to determine whether the correlation is linear.

If it is determined to be linear in step 22, proceed to step 23, where the candidate parts are narrowed down to a group of parts made of material a. If it is determined to be nonlinear, proceed to step 24, where the candidate parts are narrowed down to a group of parts made of material b. In other words, parts made of material a are parts with linear characteristics, and parts made of material b are parts with nonlinear characteristics.

The rest of the process is the same as in the previous embodiments. In step 6, a large amount of frequency data associated with vehicle speed is used to calculate the correlation between sound pressure (including harmonics) of the natural frequency of a group of parts made of material a or material b and vehicle speed using linear regression, and the regression coefficient α1 (when the regression line is y = α1x + β1) is determined. In step 4, the sound pressure-frequency spectrum associated with vehicle speed is compared with the basic sound pressure-frequency spectrum corresponding to the same vehicle speed to determine whether there is a difference between the two. If there is a difference between the two, i.e., an abnormal sound, the presence or absence of a correlation between sound pressure and vehicle speed is determined for candidate parts having a natural frequency corresponding to the frequency of the abnormal sound by determining whether the regression coefficient α1 is equal to or greater than a predetermined threshold (step 7).

If the determinations in steps 4 and 7 for a certain part A are both YES, then the part is determined to be abnormal (i.e., the source of an abnormal sound) and a message to that effect is displayed (step 8). If either result is NO, proceed to step 10, where the part is determined to be normal and a message to that effect is displayed.

In this embodiment, candidate parts are narrowed down based on whether the correlation is linear or nonlinear, making it easier to identify abnormal parts.

Figure 12 is a flowchart showing another example of the processing flow for wheel contact inspection in the fifth embodiment. In this example, as with the flowchart in Figure 11, in steps 21 and 22, it is determined whether the correlation is linear, but if it is determined in step 22 that it is not linear, the process proceeds to step 25, and a message is displayed saying "Correlated with acceleration." No particular narrowing down of candidate parts is performed.

Next, Figure 14 shows an example of the display displayed on the display unit 60. In this example display, in order to visually represent the abnormal noise caused by contact with a rotating body, the spectrogram display described above is displayed on the right side of the screen (the area indicated by the symbol A), and a two-dimensional graph of the original sound data is displayed below the spectrogram display. As mentioned above, the spectrogram is actually displayed in color. In this example display, a line-shaped portion of a different color caused by the abnormal noise appears around 5 kHz during the approximately 15-second inspection period (this is not clear in the black and white image in the attached drawing).

Also, in the upper left corner of the screen, a relatively large label indicating whether or not there is contact with the rotating body is displayed, as indicated by the symbol B. In the illustrated example, contact has been detected, so "Contact Present" is displayed in red. If no contact is detected, "No Contact" is displayed. Furthermore, if contact is present, the name of the identified part (in one example, BP: back plate) is displayed below part B in part C, the frequency of the abnormal noise (in one example, 5 kHz) is displayed in part D, the sound pressure (in one example, -30 dB) is displayed in part E, and the half-width (in one example, 100 Hz) is displayed in part F. By comparing these displayed items with the spectrogram, inspectors can make a final determination of the abnormality, i.e., whether the part is touching.

The above display is just one example; for example, the test results could be simply displayed as OK/NG or as a circle/cross symbol. It is desirable to avoid displaying too much information on the screen so that it can be understood in a short amount of time.

In addition, in the above display example, a spectrogram is used to visually represent the frequency of abnormal noise, but other displays that visualize sound, such as a mel spectrogram or keptogram, may also be used.

Next, as a different embodiment of the anomaly detection method of the present invention, the spectrogram display may be treated as image data and processed using a pre-created anomaly determination model. That is, as described above, a spectrogram display corresponding to sound data can be generated using frequency data obtained by frequency analysis of sound data. However, an anomaly determination model can be generated in advance using machine learning (or deep learning) of a large number of spectrogram display image data, including inspection examples with and without contact between various parts such as backplates. By processing the spectrogram display of the sound of the vehicle being inspected as image data using the anomaly determination model, it is possible to determine the presence or absence of abnormal sounds and identify the parts causing them based on characteristic changes in sound pressure, etc., on the image. As described above, in the completed vehicle inspection process where the vehicle speed is increased, changes over time simply correspond to changes in vehicle speed.

The above describes one embodiment in which this invention is applied to contact inspection of back plates and other parts in the wheel section during the finished vehicle inspection process, but this invention is not limited to this and can be applied in a variety of other ways. For example, it can easily inspect for contact in the wheel section during inspection and maintenance at a commercial vehicle dealership or repair shop. It is also possible to inspect for contact between parts while the vehicle is actually traveling on the road.

Note that the information presentation unit is not limited to the above-mentioned display unit that displays images, but may also present information using audio only.

REFERENCE SIGNS LIST 10: Sound measurement and acquisition unit 20: Frequency conversion unit 30: Abnormal sound determination unit 40: Natural frequency acquisition unit 50: Abnormal sound component identification unit 60: Display unit 70: Vehicle motion information acquisition unit

Claims (8)

Acquire sounds generated near the wheels of a target vehicle while changing the vehicle speed to generate sound data;
This sound data is converted according to frequency to generate frequency data associated with vehicle speed ,
This frequency data is compared with basic frequency data associated with the corresponding vehicle speed to extract abnormal noise.
The frequency characteristics of the abnormal noise are compared with the natural frequency of the part to identify the candidate part, and the correlation between the sound characteristics of the abnormal noise and the vehicle speed is used to determine whether the part in question is abnormal.
Present the results including the identified parts.
A method for detecting abnormalities in wheel sections. The method for detecting an abnormality in a wheel section according to claim 1 , wherein the sound characteristic is sound pressure. The method for detecting an abnormality in a wheel section according to claim 1 , wherein the sound characteristic is a half-width of a waveform in a frequency spectrum. 2. The method for detecting abnormalities in a wheel section according to claim 1, wherein the abnormality is determined based on a regression coefficient of a linear regression of two variables, namely , the sound pressure or half-width of the waveform in the frequency spectrum as a sound characteristic of the abnormal noise, and the vehicle speed. Extracting a fluctuating sound component having a relatively long period contained in the sound data,
The method for detecting an abnormality in a wheel section according to any one of claims 1 to 4 , wherein the frequency of the fluctuation sound component corresponds to the rotation frequency of the wheel as a weighting condition to determine that an abnormality has occurred. Determine whether the correlation between two variables, namely, the sound pressure or the half-width of the waveform in the frequency spectrum as the sound characteristic of the abnormal noise, and the vehicle speed , is linear or nonlinear;
The method for detecting an abnormality in a wheel section according to claim 1 , further comprising narrowing down candidate parts depending on whether the characteristic is linear or nonlinear. generating a spectrogram representation corresponding to the sound data using the frequency data;
A model for detecting abnormal noise is generated by machine learning using a large amount of spectrogram image data in advance.
2. The method for detecting abnormalities in a wheel section according to claim 1, further comprising the step of processing a spectrogram display of sound data to be inspected using said abnormal sound determination model to determine abnormal sounds and identify parts in an abnormal state. a sound acquisition unit that acquires sounds generated near the wheels of a target vehicle while changing the vehicle speed and generates sound data;
a frequency data generating unit that converts the sound data according to frequency to generate frequency data associated with the vehicle speed ;
an abnormal sound determination unit that extracts abnormal sounds by comparing the frequency data with basic frequency data associated with the corresponding vehicle speed ;
a part identification unit that identifies candidate parts by comparing the frequency characteristics of the abnormal noise with the natural vibration frequency of the parts, and determines whether the part is abnormal based on the correlation between the sound characteristics of the abnormal noise and the vehicle speed ;
an information presentation unit that presents information on the results including the identified parts;
A wheel section abnormality detection device comprising: