Noise factor determination device and noise factor determination method

Abstract

To provide a noise factor determination device and a noise factor determination method that can precisely determine the factor of noise.SOLUTION: A frequency analytic waveform P(x) representing variation in sound pressure level relative to a frequency is obtained with respect to sound data of an inspection sample 21 picked up when a product is driven to determine a noise factor of the inspection sample 21. The difference between a reference frequency analytic waveform Pa (x) in the same drive state as the inspection sample 21 of a reference frequency analytic waveform Pa(x) as the frequency analytic waveform P(x) of a base sample 20 prepared for each drive state of the product and an inspection frequency analytic waveform Pb(x) as the frequency analytic waveform P(x) of the inspection sample 21 is taken to obtain a noise specific waveform T(x). The factor of noise present in a measurement environment is determined from the noise specific waveform T(x).SELECTED DRAWING: Figure 3

Description

[Technical field]

[0001] The present invention relates to a noise factor determination device and a noise factor determination method for determining the cause of noise heard from a product during operation. [Background technology]

[0002] Conventionally, an abnormal sound determination device that determines whether or not an abnormal sound is included in the drive sound of a drive system product is well known (see Patent Document 1). In Patent Document 1, sound pressure levels of frequency bands are calculated for sound data of the drive sound collected by a drive sound collector. Furthermore, determination parameters are set for multiple frequency bands that are targets for abnormal sound determination, and the calculated sound pressure level is compared with the determination parameters for each frequency band. Then, the presence or absence of an abnormal sound is determined from the comparison results. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2006-126140 A Summary of the Invention [Problem to be solved by the invention]

[0004] However, drive system products do not have a constant drive speed, and the drive speed may be fast or slow depending on the set conditions. Generally, when the drive speed is fast, the sound pressure level tends to be high. Therefore, when the judgment parameter is set to a constant value regardless of the drive speed as in Patent Document 1, the drive sound may be judged as an abnormal sound even though it is a normal drive sound. Therefore, when the drive speed is relatively fast or slow, the accuracy of the abnormal sound judgment may be low.

[0005] An object of the present invention is to provide a noise factor determination device and a noise factor determination method that can determine the cause of noise with high accuracy. [Means for solving the problem]

[0006] The noise factor determination device (1) for solving the above-mentioned problems includes an analysis unit (13) that determines a frequency analysis waveform (P(x)) that indicates a change in sound pressure level relative to the frequency of sound data (Da) generated in the surrounding area when a product (2) is operating, when the sound data (Da) is collected by a sound collection unit (5); a waveform extraction unit (14) that determines a noise characteristic waveform (T(x)) by taking the difference between a reference frequency analysis waveform (Pa(x)), which is the frequency analysis waveform of a base sample (20) prepared for each operating state of the product, in the same operating state as a test sample (21), and a test frequency analysis waveform (Pb(x)), which is the frequency analysis waveform of the test sample; and a determination unit (25) that determines the cause of noise present in the measurement environment based on the noise characteristic waveform.

[0007] The noise factor determination method for solving the above-mentioned problems is a method used in a noise factor determination device (1) that determines the factors of noise present in a measurement environment when a product (2) is in operation, and causes the noise factor determination device to perform the following steps when sound data (Da) generated in the surroundings when the product is in operation is collected by a sound collection unit (5): obtain a frequency analysis waveform (P(x)) that represents the change in sound pressure level relative to the frequency of the sound data; obtain a noise characteristic waveform (T(x)) by taking the difference between a reference frequency analysis waveform (Pa(x)), which is the frequency analysis waveform of a base sample 20 prepared for each operating state of the product, in the same operating state as a test sample 21, and a test frequency analysis waveform (Pb(x)), which is the frequency analysis waveform of the test sample; and determine the factors of noise present in the measurement environment based on the noise characteristic waveform.

[0008] According to this configuration, the noise factor is determined using a reference frequency analysis waveform corresponding to the test sample. Also, the difference between the test frequency analysis waveform, which is the frequency analysis waveform of the test sample, and the test frequency analysis waveform of a base sample when the product is in the same operating state as when the test sample is measured is calculated, and the noise factor is determined by analyzing the noise specific waveform, which is the difference in these waveforms. In this way, by using the optimal reference frequency analysis waveform, it is possible to appropriately distinguish the noise factor from the waveform change of the noise specific waveform, which is the difference between the reference frequency analysis waveform and the test frequency analysis waveform. Therefore, the noise factor can be determined with high accuracy.

[0009] Note that, by way of definition, the same operating state does not necessarily mean that, when there are multiple settings for operating a product, all of the settings are the same, but also includes a state in which some of the settings are different. [Brief description of the drawings]

[0010] [Figure 1] 1 is a configuration diagram of a noise cause determination device according to an embodiment; [Diagram 2] FIG. 11 is a waveform diagram of a frequency analysis waveform. [Diagram 3] FIG. 4 is a waveform diagram of a noise specific waveform. [Figure 4] FIG. 2 is a schematic diagram of structured data. [Diagram 5] FIG. 11 is an explanatory diagram showing a method for correcting a noise specific waveform. [Figure 6] FIG. 1 is a schematic diagram of noise cause determination using a learning model. [Figure 7] FIG. 2 is an explanatory diagram showing a structured data group used as learning data. [Figure 8] FIG. 1 is a schematic diagram illustrating how sound data of a test sample is collected by a mobile device. [Figure 9] FIG. 13 is a peak distribution diagram of a test sample analyzed by a learning model. [Figure 10] FIG. 13 is a model diagram showing the determination process of the determination results of the learning model. [Figure 11] 10A and 10B are explanatory diagrams showing an example in which erroneous determination occurs due to fluctuations in characteristic values; [Figure 12] FIG. 11 is a procedure diagram for issuing an instruction to change the drive conditions or assembly state and identifying the final noise factor. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0011] Hereinafter, one embodiment of the present disclosure will be described. [Overall configuration] As shown in Fig. 1, a noise factor determination device 1 determines the factor of noise generated in a measurement environment when a product 2 is operated. The product 2 is, for example, an in-vehicle product. An example of an in-vehicle product is a blower that blows air into the interior of a vehicle. In addition, examples of the factors of the noise include vibration of the product 2, sliding sounds and friction sounds of parts of the product 2, etc.

[0012] [Prerequisite configuration of noise factor determination device 1] The noise factor determination device 1 is provided in a mobile device 3 that is operated by an operator when determining noise factors. The mobile device 3 is, for example, a notebook personal computer or a dedicated tool. The mobile device 3 includes a control unit 4, a sound collection unit 5, an operation unit 6, and a display unit 7. The control unit 4 is constructed from devices such as a CPU (Central Processing Unit), ROM (Read Only Memory), and RAM (Random Access Memory). The control unit 4 controls the operation of the mobile device 3.

[0013] The sound collection unit 5 is connected to the control unit 4 via the interface 8 of the mobile device 3. The sound collection unit 5 is, for example, a microphone, and outputs collected sound data to the control unit 4 via the interface 8. In this example, the sound collection unit 5 collects sound data Da output from the product 2 when it is driven, and outputs it to the control unit 4.

[0014] The operation unit 6 is connected to the control unit 4 via an interface 9 of the mobile device 3. The operation unit 6 is, for example, a numeric keypad, a keyboard, or a mouse of the mobile device 3. The operation unit 6 outputs an operation signal to the control unit 4 via the interface 9 when the operation unit 6 is operated.

[0015] The display unit 7 is connected to the control unit 4 via an interface 10 of the mobile device 3. The display unit 7 is, for example, a display. When an image signal is input from the control unit 4 via the interface 10, the display unit 7 displays the image signal on a screen.

[0016] [Specific configuration of noise factor determination device 1] The noise factor determination device 1 includes an analysis unit 13, a waveform extraction unit 14, a waveform output processing unit 15, a correction unit 16, and an operation instruction unit 17. The analysis unit 13, the waveform extraction unit 14, the waveform output processing unit 15, the correction unit 16, and the operation instruction unit 17 are provided in the control unit 4.

[0017] As shown in Fig. 2, when sound data Da output from the product 2 during operation is collected by the sound collection unit 5, the analysis unit 13 obtains a frequency analysis waveform P(x) that represents the change in sound pressure level relative to the frequency of the sound data Da. The frequency analysis waveform P(x) is a "frequency-sound pressure level" change characteristic that represents the level of sound pressure level generated at each frequency. In general, when the frequency is low, the sound pressure level tends to be high, and as the frequency increases, the sound pressure level decreases.

[0018] The waveform output processing unit 15 outputs the frequency analysis waveform P(x) determined by the analysis unit 13 to the display unit 7. In this example, it is preferable that the waveform output processing unit 15 displays the waveform of the frequency analysis waveform P(x) itself on the display unit 7.

[0019] 3, the waveform extraction unit 14 obtains the noise specific waveform T(x) by taking the difference between the reference frequency analysis waveform Pa(x) in the same driving state as the inspection sample 21 and the inspection frequency analysis waveform Pb(x) which is the frequency analysis waveform P(x) of the inspection sample 21, out of the reference frequency analysis waveform Pa(x) which is the frequency analysis waveform P(x) of the base sample 20 prepared for each driving state of the product 2. Examples of the driving state of the product 2 include the on / off state of a switch provided in the product 2, the driving speed of an actuator of the product 2, and the driving mode set for the product 2. If the inspection sample 21 contains a noise component, a peak waveform 22 appears at a predetermined position of the noise specific waveform T(x).

[0020] Note that the "same driving state" does not necessarily mean that all the settings of the product 2 are the same. For example, as described above, it includes partial differences in settings, such as [1] the on / off state of the switch of the product 2, [2] the driving speed of the actuator of the product 2, and [3] the driving mode of the product 2, where [1] and [2] are the same and only [3] is different. In other words, it also includes differences in any one of [1] to [3]. For example, in the case of a blower, the on / off state of the switch is the on / off of the blower switch. For example, in the case of a blower, the driving speed of the actuator is the adjustment of the air volume. For example, in the case of a blower, the driving mode is the setting of the air outlet (foot, face, def, etc.) and the on / off of the temperature adjustment function.

[0021] 4, the waveform extraction unit 14 converts the raw data of the noise specific waveform T(x) into structured data Dk by extracting a characteristic value Kx corresponding to the difference between the reference frequency analysis waveform Pa(x) and the test frequency analysis waveform Pb(x) for each divided frequency range Ef. The characteristic value Kx is a value based on the prominence amount H of the peak waveform 22 detected in each frequency range Ef. The peak waveform 22 is a point where the prominence amount H is equal to or greater than the specified value Hbs and has the largest value within the frequency range Ef.

[0022] Each frequency domain Ef is set to have a non-uniform range so as to have a range according to the type of noise expected in the measurement environment. In FIG. 4, "Fa", "Fb", "Fc", "Fd", "Fe", and "Ff" are illustrated. As a method of setting the frequency domain Ef to be non-uniform, for example, "Fa" is set to a 1 kHz range, the adjacent "Fb" is set to a 4 kHz range, and the further adjacent "Fc" is set to a 2 kHz range.

[0023] 5, the correction unit 16 corrects the noise specific waveform T(x) in the sound pressure level axis direction. In this example, it is preferable that the correction unit 16 performs the correction by calculating an average value Tavr of the sound pressure level of the noise specific waveform T(x) and offsetting the noise specific waveform T(x) in the sound pressure level axis direction (specifically, the vertical axis direction) by the value of the average value Tavr.

[0024] 1, the noise factor determination device 1 includes a determination unit 25 that determines the factor of noise present in the measurement environment when the product 2 is operated. The determination unit 25 is provided in the control unit 4. The determination unit 25 determines the factor of the noise based on the specific noise waveform T(x) (structured data Dk) obtained by the waveform extraction unit 14.

[0025] [Configuration that uses learning to determine noise factors] The noise factor determination device 1 includes a learning model 28 and a determination process output unit 29. The determination unit 25 determines the cause of the noise using the learning model 28 generated by learning past sound data Da. The learning model 28 is written and stored in the memory 30 of the mobile device 3. It is preferable that the learning model 28 is updated to a new model periodically, for example. The determination process output unit 29 is provided in the control unit 4. The determination process output unit 29 outputs to the display unit 7 the determination process when the learning model 28 is used to determine the noise factor.

[0026] 6, the learning model 28 is generated by using the feature quantities of structured data Dk (learning structured data Dk1) generated based on sound data Da0 of a non-defective product and sound data Da1 of a past defective product, and learning these feature quantities. The learning structured data Dk1 is generated by preprocessing the sound data Da1 of a past defective product using the sound data Da0 of a non-defective product. The preprocessing is preferably a process of, for example, determining the specific noise waveforms T(x) of each defective product using the sound data Da0 of a non-defective product as a base sample 20, and generating learning structured data Dk1 of these specific noise waveforms T(x).

[0027] In this example, the learning automatically creates a decision threshold Rpk for each frequency domain Ef used to distinguish the presence or absence of a peak in the sound data Da2 of the test sample 21. In this way, the decision threshold Rpk is automatically set for each frequency domain Ef by learning. The decision threshold Rpk is used when determining whether the characteristic value Kx for each frequency domain Ef is set to whether there is a peak or not in the structured test data Dk2 of the test sample 21. The generated learning model 28 is written and stored in the memory 30 of the mobile device 3.

[0028] When determining the cause of noise in the inspection sample 21, the noise cause determination device 1 inputs sound data Da2 of the inspection sample 21 via the interface 8. This sound data Da2 is converted into structured data Dk (structured data for inspection Dk2) by preprocessing in the analysis unit 13 and waveform extraction unit 14. The preprocessing is preferably a process of determining the noise specific waveform T(x) of the inspection sample 21 using sound data Da0 of a non-defective product as the base sample 20, and generating structured data for inspection Dk2 of this noise specific waveform T(x).

[0029] The determination unit 25 determines the cause of noise in the test sample 21 by analyzing the structured data for testing Dk2 of the test sample 21 using the learning model 28 (determination threshold Rpk). In the structured data for testing Dk2 of the test sample 21, in each frequency domain Ef, a portion where the characteristic value Kx is equal to or greater than the determination threshold Rpk is determined to be "peak-present," and a portion where the characteristic value Kx is less than the determination threshold Rpk is determined to be "peak-less." The determination unit 25 outputs an image signal related to this determination result to the display unit 7 via the interface 10, thereby displaying the determination result of the noise cause on the screen of the display unit 7.

[0030] Next, the operation of the noise factor determination device 1 (noise factor determination method) of this embodiment will be described. [Construction of learning model 28] 7, in constructing the learning model 28, first, raw data (unstructured data) of a specific noise waveform T(x) generated from sound data Da2 of a past defective product is converted into structured data Dk, which is then stored as learning structured data Dk1. The learning structured data Dk1 is data obtained by organizing the raw data (unstructured data) into a predetermined format, and in this example, represents the distribution of the characteristic value Kx for each frequency range Ef.

[0031] The learning structured data Dk1 is generated by converting a noise specific waveform T(x) obtained from the difference between a frequency analysis waveform P(x) of past sound data Da1 of a defective product and a frequency analysis waveform P(x) of a base sample 20. In this example, for example, sound data Da0 of a non-defective product is used as the base sample 20. In addition, for the base sample 20, a sample obtained when the product is in the same driving state (driving state of product 2) as when the sound data Da1 of the defective product was measured during learning is used.

[0032] The structured learning data Dk1 is compiled as a distribution table 33 that lists the structured learning data Dk1. The vertical axis of the distribution table 33 is an ID column in which the ID number of each test sample 21 is input. The horizontal axis of the distribution table 33 includes a parameter input column 34 in which a characteristic value Kx is input for each frequency range Ef, and a factor write column 35 in which a noise factor is written. In this example, for "ID1" written in the distribution table 33, for convenience, the data of the base sample 20 is written, and the characteristic values ​​Kx in all parameter columns are set to "1".

[0033] For "ID2" in the learning structured data Dk1, the characteristic value Kx is "2" in the frequency domains "Fd" and "Fg", and "5" in the frequency domains "Fe" and "Ff". The remaining frequency domain Ef has a characteristic value Kx of "1", that is, a value equivalent to the base sample 20 (no peak waveform 22). For "ID2" in the learning structured data Dk1, the noise factor is set to "Factor A". In other words, when the characteristic value Kx is distributed as in "ID2", the noise factor is identified as "Factor A". Incidentally, the noise factor for "ID3" is set as "Factor B", the noise factor for "ID4" is set as "Factor C", and the noise factor for "ID5" is set as "Factor D".

[0034] The learning model 28 is generated by learning a data group of the learning structured data Dk1 summarized in a distribution table 33 as learning data. Specifically, features are extracted from each of the learning structured data Dk1, and the features are analyzed to calculate an optimal learning model 28. Various learning methods can be used, such as random forest, decision tree, support vector machine, linear regression, and logistic regression.

[0035] In this example, the generation of the learning model 28 is a process of automatically generating a judgment threshold Rpk for setting the presence or absence of a peak for the characteristic value Kx of each frequency domain Ef of the structured data for inspection Dk2 of the inspection sample 21. For example, in the case of the frequency domain "Fe", the judgment threshold Rpk of the frequency domain "Fe" is automatically generated by performing learning using the characteristic value Kx "5" of "ID2", the characteristic value Kx "5" of "ID3", the characteristic value Kx "1" of "ID4", and the characteristic value Kx "1" of "ID5" as indices (ID5 and subsequent values ​​are omitted). A similar process is performed for the other frequency domains Ef, so that individual judgment thresholds Rpk are automatically set for each frequency domain Ef. These judgment thresholds Rpk are written and stored in the memory 30 as the learning model 28.

[0036] [Noise factor determination] As shown in Fig. 8, when it is desired to determine the cause of noise in sound data Da2 of a test sample 21, a sound collection operation is performed on the operation unit 6 of the mobile device 3, whereby the sound data Da2 of the test sample 21 is collected by the sound collection unit 5 of the mobile device 3. The collected sound data Da2 is input to the control unit 4 via the interface 8. The sound data Da2 input to the control unit 4 is preprocessed by the control unit 4. The preprocessing is a process of converting the collected sound data Da2 into structured data for testing Dk2.

[0037] The pre-processing will be explained below. First, the analysis unit 13 obtains a frequency analysis waveform P(x) that indicates a change in sound pressure level with respect to the frequency of the sound data Da2 (see FIG. 2). Specifically, the analysis unit 13 obtains an inspection frequency analysis waveform Pb(x), which is the frequency analysis waveform P(x) of the inspection sample 21, from the collected sound data Da2 of the inspection sample 21.

[0038] Next, the waveform extraction unit 14 prepares a base sample 20 under the same product driving conditions (driving conditions of product 2) as this inspection sample 21. Then, the waveform extraction unit 14 extracts the noise characteristic waveform T(x) of this inspection sample 21 by taking the difference between the above-mentioned inspection frequency analysis waveform Pb(x) and the frequency analysis waveform P(x) (reference frequency analysis waveform Pa(x)) of the prepared base sample 20 (see FIG. 3).

[0039] The correction unit 16 preferably corrects the noise specific waveform T(x) of the test sample 21 (see FIG. 5). When making the correction, the correction unit 16 first calculates the average value Tavr of the sound pressure level (waveform amplitude) of the noise specific waveform T(x). Next, the correction unit 16 offsets the noise specific waveform T(x) in the vertical axis direction (sound pressure level axis direction) by the value of the average value Tavr to correct the position to the zero point on the vertical axis. For example, if the average value Tavr is a positive value, the correction unit 16 subtracts the average value Tavr from the noise specific waveform T(x) in the vertical axis direction. On the other hand, if the average value Tavr is a negative value, the correction unit 16 adds the average value Tavr to the noise specific waveform T(x) in the vertical axis direction.

[0040] After extracting the particular noise waveform T(x), the waveform extraction unit 14 extracts a characteristic value Kx corresponding to the waveform of the particular noise waveform T(x) for each divided frequency range Ef, thereby converting the raw data of the particular noise waveform T(x) into structured test data Dk2 (see FIG. 4). In this example, the difference between the reference frequency analysis waveform Pa(x) and the test frequency analysis waveform Pb(x) for each frequency range Ef is quantified as the characteristic value Kx. The larger the characteristic value Kx, the greater the waveform difference.

[0041] 9, after generating the structured data for inspection Dk2, the judgment unit 25 judges the noise factor of the inspection sample 21 by inputting the generated structured data for inspection Dk2 to the learning model 28. Specifically, the judgment unit 25 compares the characteristic value Kx and the judgment threshold Rpk in each frequency domain Ef of the structured data for inspection Dk2 of the inspection sample 21, and judges the frequency domain Ef in which the characteristic value Kx is equal to or greater than the judgment threshold Rpk as "peak-present" and the frequency domain Ef in which the characteristic value Kx is less than the judgment threshold Rpk as "peak-less." In the example shown in the figure, the frequency domains "Fe," "Ff," and "Fj" are judged as "peak-present," and the rest are judged as "peak-less."

[0042] The determination unit 25 identifies the noise factors using the structured data for inspection Dk2 for which the presence or absence of peaks has been determined. Specifically, in the structured data for inspection Dk2 for which the presence or absence of peaks has been determined, "no peak" is set to a characteristic value of "1" and the region for "peak presence" is set to a characteristic value of "2 or more", and the noise factors are identified from the combination of these characteristic values ​​Kx. In this example, since peaks are present in the frequency regions "Fe", "Ff", and "Fj", the noise factor (factor B) associated with "ID3" which is similar to this parameter combination is identified as the noise factor of the inspection sample 21.

[0043] The determination unit 25 displays the data of the determination result on the screen of the display unit 7 via the interface 10. The operator can determine the noise factor of the test sample 21 by checking the determination result displayed on the display unit 7.

[0044] [Visualization of the decision-making process leading to a decision result] 10, the judgment process output unit 29 outputs the judgment process when noise judgment is performed using the learning model 28 to the display unit 7. In this example, the judgment process is displayed on the display unit 7 as a judgment process display screen 38, which represents the judgment process leading up to the judgment result in a format that allows visual confirmation. The judgment process output unit 29 also displays the judgment process leading up to the judgment result on the display unit 7 in a format in which the judgment branches are represented in a tree shape.

[0045] As shown in the figure, the judgment process display screen 38 includes a data display 39 of the structured data Dk (structured data for inspection Dk2) of the inspection sample 21 to be confirmed, and an analytical model 40 which graphically models the judgment process. The operator checks the process up to when the judgment unit 25 issues a judgment result on the noise cause by comparing the data display 39 with the analytical model 40. The data display 39 lists the characteristic values ​​Kx of the structured data for inspection Dk2.

[0046] The analysis model 40 is a model of the judgment result when the comparison between the characteristic value Kx and the judgment threshold value Rpk is made in sequence while changing the frequency domain Ef. The judgment unit 25 tries to efficiently classify the noise factors (factors A to F in the figure). For this reason, as shown in the figure, there are frequency domains Ef (frequency domains Fa, Fd) that are not used for judging the noise factors, and there are multiple frequency domains Ef (frequency domains Fc).

[0047] In this example, in the first stage of the analysis model 40, it is confirmed whether or not the characteristic value Kx in the frequency domain Ff is equal to or less than the judgment threshold Rpk (step 101). If the characteristic value Kx in the frequency domain Ff is equal to or less than the judgment threshold Rpk, the process proceeds to step 102, and if the characteristic value Kx in the frequency domain Ff exceeds the judgment threshold Rpk, the process proceeds to step 103. In the case of the analysis model 40 of this example, the characteristic value Kx in the frequency domain Ff is equal to or less than the judgment threshold Rpk, so the process proceeds to step 102.

[0048] In the second stage of the analysis model 40, it is confirmed whether the characteristic value Kx is equal to or less than the judgment threshold value Rpk in the frequency domain Fb (step 102). If the characteristic value Kx is equal to or less than the judgment threshold value Rpk in the frequency domain Fb, the process proceeds to step 104, and if the characteristic value Kx exceeds the judgment threshold value Rpk in the frequency domain Fb, the process proceeds to step 105.

[0049] In the third stage of the analysis model 40, it is confirmed whether the characteristic value Kx is equal to or less than the judgment threshold value Rpk in the frequency domain Fc (step 104). If the characteristic value Kx is equal to or less than the judgment threshold value Rpk in the frequency domain Fc, a judgment result of "no abnormality" is derived, and if the characteristic value Kx exceeds the judgment threshold value Rpk in the frequency domain Fc, a judgment result that the noise factor is "Factor A" is derived. In this way, the judgment process leading to the judgment result of the noise factor can be confirmed by following the table or diagram.

[0050] [Stratified analysis of noise factors] As shown in FIG. 11, when noise is generated in a product 2 ("ID13" in the figure) during driving, the characteristic value Kx of the frequency domain Ef where the peak waveform 22 occurs varies depending on the number and size of foreign objects present in the drive system of the product 2. As shown in the example of the figure, in the case of the frequency domain "Fd" of noise caused by foreign objects, the characteristic value Kx of "Fd" varies depending on the number and size of foreign objects. In other words, if the foreign object is small and the characteristic value Kx is a low value (for example, "2"), "Fd" may be judged as "no peak" depending on the value of the automatically generated judgment threshold value Rpk. In this case, although the correct judgment result is "Factor F," "Factor F" linked to "ID12" with a similar peak distribution is output as the judgment result, resulting in an erroneous judgment.

[0051] Therefore, as shown in FIG. 12, when there is another structured data Dk (any of the learning structured data Dk1) similar to the inspection structured data Dk2 of the inspection sample 21, the operation instruction unit 17 performs a stratified analysis of the noise factors. In this example, the operation instruction unit 17 issues an instruction to the worker to change the operating conditions or assembly state of the product 2. This instruction is preferably executed, for example, via the display unit 7. A series of processes for determining the noise factors is executed in the same manner after the operating conditions or assembly state are changed. The operation instruction unit 17 repeatedly executes the instruction to change the operating conditions or assembly state with different instructions until the noise factor is identified.

[0052] The stratified analysis of noise factors will be specifically described. First, in step 201, the operation instruction unit 17 judges whether or not there is another data (learning structured data Dk1) similar to the inspection structured data Dk2 of the inspection sample 21. In this example, the operation instruction unit 17 checks whether or not there is another learning structured data Dk1 having a similar distribution to the numerical distribution of the characteristic value Kx of the inspection sample 21 based on the numerical distribution of the characteristic value Kx of the inspection sample 21. For example, if the total number of frequency regions Ef in which the characteristic value Kx matches is equal to or greater than a predetermined number, it is preferable to judge that similar other data exists. If similar other data does not exist, the process proceeds to step 202, and if similar other data exists, the process proceeds to step 203.

[0053] If there is no similar data in step 202, the operation instruction unit 17 causes the determination unit 25 to determine the noise factor in the normal manner. Therefore, the determination unit 25 executes the determination of the noise factor according to the normal procedure.

[0054] In step 203, if similar other data exists, the operation instruction unit 17 outputs an instruction to change the driving conditions or assembly state of the product 2. Examples of changing the driving conditions include turning a switch on and off, changing the driving speed of the product 2, or changing the driving mode of the product 2. Examples of changing the assembly state include disassembling the product 2 or removing a specific part.

[0055] In step 204, the operation instruction unit 17 causes the judgment unit 25 to judge the noise factor again. For this reason, the judgment unit 25 executes judgment again according to the normal procedure. Furthermore, when the assembly state of the product 2 is changed, the operator may be caused to visually check the noise factor of the product 2 whose assembly state has been changed.

[0056] In step 205, the operation instruction unit 17 judges whether the noise factor has been identified after the noise factor re-determination has been performed. If the noise factor has been identified, the stratified analysis process ends. On the other hand, if the factor cannot be identified, the process returns to step 203, and an instruction is given to change the operating conditions or assembly state to re-determine the noise factor. The above series of processes are then repeated until the noise factor can be determined.

[0057] According to the noise factor determination device 1 (noise factor determination method) of the above embodiment, the following effects can be obtained. (1) When sound data Da generated in the surroundings while the product 2 is in operation is collected by the sound collection unit 5, the analysis unit 13 of the noise factor determination device 1 obtains a frequency analysis waveform P(x) that indicates the change in sound pressure level relative to the frequency of the sound data Da. The waveform extraction unit 14 of the noise factor determination device 1 obtains a noise specific waveform T(x) by taking the difference between a reference frequency analysis waveform Pa(x) in the same operating state as the test sample 21 and a test frequency analysis waveform Pb(x) that is the frequency analysis waveform P(x) of the test sample 21, among the reference frequency analysis waveforms Pa(x) that are the frequency analysis waveforms P(x) of the base samples 20 prepared for each operating state of the product 2. The determination unit 25 of the noise factor determination device 1 determines the cause of the noise present in the measurement environment based on the noise specific waveform T(x).

[0058] According to the configuration of this example, the noise factor is determined using the reference frequency analysis waveform Pa(x) corresponding to the inspection sample 21. Also, the difference between the inspection frequency analysis waveform Pb(x), which is the frequency analysis waveform P(x) of the inspection sample 21, and the reference frequency analysis waveform Pa(x) of the base sample 20 when the product 2 is in the same driving state as when the inspection sample 21 is measured, is calculated, and the noise factor is determined by analyzing the noise characteristic waveform T(x), which is the difference in waveforms. In this way, the optimal reference frequency analysis waveform Pa(x) is used, and it is possible to appropriately distinguish the noise factor from the waveform change of the noise characteristic waveform T(x), which is the difference between the reference frequency analysis waveform Pa(x) and the inspection frequency analysis waveform Pb(x). Therefore, the noise factor can be determined with high accuracy.

[0059] (2) The correction unit 16 of the noise factor determination device 1 corrects the noise specific waveform T(x) in the direction of the sound pressure level axis. With this configuration, it is possible to optimize the position of the noise specific waveform T(x) in the direction of the sound pressure level axis by correction. This makes it difficult for a situation to occur in which there is noise when there is no noise, or there is noise when there is no noise.

[0060] (3) The waveform extraction unit 14 converts the raw data of the noise specific waveform T(x) into structured data Dk by extracting a characteristic value Kx corresponding to the difference between the reference frequency analysis waveform Pa(x) and the test frequency analysis waveform Pb(x) for each divided frequency range Ef. The determination unit 25 determines the cause of the noise based on the structured data Dk. With this configuration, the noise cause is determined using the structured data Dk in which the characteristic value Kx calculated as the value corresponding to the difference between the reference frequency analysis waveform Pa(x) and the test frequency analysis waveform Pb(x) for each frequency range Ef is organized into a specific form, so that the determination process does not become complicated.

[0061] (4) The characteristic value Kx is a value based on the prominence H of the peak waveform 22 detected in each frequency range Ef. With this configuration, the cause of a loud noise can be accurately determined and the noise can be dealt with.

[0062] (5) The peak waveform 22 is a portion where the protrusion amount H is equal to or greater than the specified value Hbs and has the largest value within the frequency domain Ef. With this configuration, it is possible to identify one specified portion within the frequency domain Ef as the peak waveform 22. Therefore, a situation where multiple peak waveforms 22 are generated does not occur.

[0063] (6) Each frequency range Ef is set to have a non-uniform range according to the type of noise expected in the measurement environment. With this configuration, the frequency range Ef is associated with the noise factor, so that if the frequency range Ef in which the peculiar characteristic value Kx occurs in the test sample 21 is identified, the noise factor can be identified. This contributes to improving the accuracy of determining the noise factor.

[0064] (7) If there is another structured data Dk similar to the structured data Dk of the inspection sample 21, the operation instruction unit 17 of the noise factor determination device 1 instructs the worker to change the operating conditions or assembly state of the product 2. A series of processes for determining the cause of the noise is executed in the same manner after the operating conditions or assembly state are changed. The operation instruction unit 17 repeatedly executes the instruction to change the operating conditions or assembly state with different instructions until the cause of the noise is identified. With this configuration, even if there is another structured data Dk similar to the structured data Dk of the inspection sample 21, it is possible to narrow down the noise factors through re-determination by changing the operating conditions or assembly state. This is more advantageous in identifying the correct noise factor.

[0065] (8) The determination unit 25 determines the cause of noise by a learning model 28 generated by learning past sound data Da. With this configuration, since the noise cause is determined by learning, it is possible to reduce the variation in determination compared to when the noise cause is identified by a person. This contributes to improving the accuracy of determining the noise cause.

[0066] (9) The learning model 28 is generated by using the features of the structured data Dk (learning structured data Dk1) generated based on the sound data Da0 of a non-defective product and the sound data Da1 of a past defective product, and learning these features. With this configuration, it is possible to execute learning using the structured data Dk (learning structured data Dk1) in which the specific noise waveform T(x) is organized into a specific form as learning data. This makes it easy to handle the learning data, and therefore does not complicate the task of generating the learning model 28.

[0067] (10) In the learning, a judgment threshold Rpk used to distinguish the presence or absence of a peak in the sound data Da2 of the test sample 21 is automatically created for each frequency domain Ef. With this configuration, it is possible to automatically generate the judgment threshold Rpk by learning, so that it is possible to set the judgment threshold Rpk optimal for judging the noise factor for each frequency domain Ef. This contributes to improving the accuracy of judging the noise factor.

[0068] (11) The judgment process output unit 29 of the noise factor judgment device 1 outputs the judgment process when the learning model 28 judges the noise factors to the display unit 7. This configuration enables the operator to check the judgment process leading up to the judgment result. This makes it possible to immediately address any doubts about the noise factor judgment content or to correct the judgment process program.

[0069] (12) The judgment process output unit 29 displays the judgment process leading to the judgment result in a format in which the branches of judgment are represented in a tree shape on the display unit 7. According to this configuration, the judgment process leading to the judgment result can be notified to the operator in an easy-to-read tree format.

[0070] (13) The waveform output processing unit 15 of the noise factor determination device 1 outputs the frequency analysis waveform P(x) determined by the analysis unit 13 to the display unit 7. This configuration makes it possible to directly check the frequency analysis waveform P(x) visually, which can be used as a clue for identifying the noise factor. This contributes to improving the accuracy of noise factor determination.

[0071] This embodiment can be modified as follows: This embodiment and the following modifications can be combined with each other to the extent that no technical contradiction occurs. The learning sound data Da (Da0, Da1) is not limited to data collected by the sound collection unit 5, and sound data Da measured by another device may be imported from the outside as data.

[0072] It is preferable that the correction unit 16 also corrects the specific noise waveform T(x) of the past sound data Da1 of a defective product used in learning. Each frequency range Ef does not necessarily have to be set so that adjacent ranges are continuous with each other, and may be set so that adjacent ranges are discontinuous with each other.

[0073] The sound collector 5 is not limited to being a member attached to the mobile device 3, and may be a separate member from the mobile device 3. The method of correcting the noise characteristic waveform T(x) is not limited to the method of finding the average value Tavr and offsetting the waveform in the axial direction. For example, other methods may be used, such as correcting only the peak value of the waveform to an optimal value.

[0074] The value of the characteristic value Kx is not limited to a value corresponding to the protrusion amount H of the peak waveform 22. For example, it may be a binary value representing the presence or absence of a peak, or it may be a value that increases exponentially as the waveform difference becomes larger.

[0075] The characteristic value Kx is not limited to the peak value of the waveform, but may be, for example, the average value or the number of peaks, or a composite value of these. The display of the determination process is not limited to the tree-like format shown in FIG. 10, but may be changed to another format, such as a list of explanatory text.

[0076] After the learning model 28 is generated, it may be evaluated using test data. Examples of evaluation methods include hold-out methods and cross-validation methods. The learning model 28 may be generated, for example, by deep learning (neural networks).

[0077] The method for determining the noise factor is not limited to the method using the learning model 28. In other words, the noise factor may be determined simply from the waveform difference between the base sample 20 and the test sample 21 without using the learning model 28.

[0078] The analysis unit 13, the waveform extraction unit 14, the waveform output processing unit 15, the correction unit 16, the operation instruction unit 17, the judgment unit 25, and the judgment process output unit 29 may be configured as follows: [1] one or more processors that operate according to a computer program (software), or [2] a combination of such a processor and one or more dedicated hardware circuits such as an application specific integrated circuit (ASIC) that executes at least some of the various processes. The processor includes a CPU and a memory such as a RAM and a ROM, and the memory stores program code or instructions that are configured to cause the CPU to execute the processes. The memory (computer-readable medium) includes any available medium that can be accessed by a general-purpose or dedicated computer. Alternatively, instead of a computer including the above processor, a processing circuit configured by one or more dedicated hardware circuits that execute all of the various processes may be used.

[0079] The analysis unit 13, the waveform extraction unit 14, the waveform output processing unit 15, the correction unit 16, the operation instruction unit 17, the judgment unit 25, and the judgment process output unit 29 may be configured from independent processors, or may be constructed from a processor that shares some of its functions. In this way, the analysis unit 13, the waveform extraction unit 14, the waveform output processing unit 15, the correction unit 16, the operation instruction unit 17, the judgment unit 25, and the judgment process output unit 29 are not limited to being independent functional blocks, and may be configured from one functional block, or may be configured from a functional block that shares some of its functions.

[0080] Although the present disclosure has been described with reference to the embodiment, it is understood that the present disclosure is not limited to the embodiment or structure. The present disclosure also includes various modifications and modifications within the scope of equivalents. In addition, various combinations and forms, as well as other combinations and forms including only one element, more than one element, or less than one element, are also within the scope and concept of the present disclosure. [Explanation of symbols]

[0081] 1...noise factor judgment device, 2...product, 5...sound collection unit, 7...display unit, 13...analysis unit, 14...waveform extraction unit, 15...waveform output processing unit, 16...correction unit, 17...operation instruction unit, 20...base sample, 21...test sample, 22...peak waveform, 25...judgment unit, 28...learning model, 29...judgment process output unit, Da...sound data, P(x)...frequency analysis waveform, Pa(x)...reference frequency analysis waveform, Pb(x)...test frequency analysis waveform, T(x)...noise specific waveform, Ef...frequency domain, Px...characteristic value, Dk...structured data, H...prominence amount, Hbs...specified amount, judgment threshold...Rpk.

Claims

1. an analysis unit (13) for obtaining a frequency analysis waveform (P(x)) which represents a change in sound pressure level with respect to the frequency of sound data collected by a sound collection unit (5) when sound data (Da) generated in the surroundings during operation of the product (2) is collected; a waveform extraction unit (14) for determining a noise characteristic waveform (T(x)) by calculating a difference between a reference frequency analysis waveform (Pa(x)) which is the frequency analysis waveform of a base sample (20) prepared for each operating state of the product, in the same operating state as a test sample (21), and a test frequency analysis waveform (Pb(x)) which is the frequency analysis waveform of the test sample; a determination unit (25) for determining a cause of noise present in the measurement environment based on the noise characteristic waveform; Equipped with the waveform extraction unit converts the raw data of the noise specific waveform into structured data (Dk) by extracting a characteristic value (Px) corresponding to a difference between the reference frequency analysis waveform and the test frequency analysis waveform for each divided frequency range (Ef); The determination unit determines a cause of the noise based on the structured data, an operation instruction unit (17) that issues an instruction to an operator to change a driving condition or an assembly state of the product when another structured data similar to the structured data of the inspection sample exists; A series of processes for determining the cause of noise is similarly executed after the driving conditions or the assembly state are changed, The operation instruction unit repeatedly executes instructions to change the drive conditions or the assembly state by changing the instructions until the cause of the noise is identified.

2. A correction unit (16) is provided for correcting the noise characteristic waveform in the axial direction of the sound pressure level. The noise cause determining device according to claim 1.

3. The characteristic value is a value based on the protrusion amount (H) of the peak waveform (22) detected in each of the frequency ranges. The noise factor determination device according to claim 1 or 2.

4. The peak waveform is a point where the amount of protrusion is equal to or greater than a specified value (Hbs) and where the value is the largest within the frequency domain. The noise cause determining device according to claim 3.

5. The frequency ranges are set non-uniformly so as to correspond to the types of noise expected in the measurement environment. The noise factor determination device according to any one of claims 1 to 4.

6. The determination unit determines the cause of noise using a learning model (28) generated by learning the past sound data. The noise factor determination device according to any one of claims 1 to 5.

7. The learning model is generated by using features of the structured data generated based on the sound data of a non-defective product and the sound data of a defective product in the past, and learning the features. The noise cause determining device according to claim 6.

8. The learning automatically creates a decision threshold (Rpk) for each frequency region, which is used to distinguish the presence or absence of a peak in the sound data of the test sample. The noise cause determining device according to claim 7.

9. A judgment process output unit (29) is provided for outputting the judgment process when the noise factor is judged by the learning model to a display unit (7). The noise factor determination device according to any one of claims 6 to 8.

10. The judgment process output unit displays the judgment process up to the judgment result on the display unit in a format in which the branches of judgment are represented in a tree form. The noise cause determining device according to claim 9.

11. a waveform output processing unit (15) for outputting the frequency analysis waveform obtained by the analysis unit to a display unit; The noise factor determination device according to any one of claims 1 to 10.

12. A noise factor determination method used in a noise factor determination device (1) that determines the factors of noise present in a measurement environment when a product (2) is in operation, comprising: When sound data (Da) generated in the surroundings during operation of the product is collected by a sound collecting unit (5), a frequency analysis waveform (P(x)) is obtained that represents a change in sound pressure level with respect to the frequency of the sound data; determining a noise characteristic waveform (T(x)) by taking a difference between a reference frequency analysis waveform (Pa(x)) which is the frequency analysis waveform of a base sample (20) prepared for each operating state of the product, in the same operating state as a test sample (21), and a test frequency analysis waveform (Pb(x)) which is the frequency analysis waveform of the test sample; extracting a characteristic value (Px) corresponding to a difference between the reference frequency analysis waveform and the test frequency analysis waveform for each divided frequency range (Ef), thereby converting the raw data of the noise specific waveform into structured data (Dk); determining a cause of noise present in a measurement environment based on the structured data; and When another structured data similar to the structured data of the inspection sample exists, an instruction is given to a worker to change a driving condition or an assembly state of the product. The noise factor determination device executes the above-mentioned a series of processes for determining the cause of noise is executed by the noise cause determination device after the driving condition or the assembly state is changed; The noise factor determination method includes causing the noise factor determination device to repeatedly execute instructions to change the driving conditions or the assembly state with different instructions changed until the cause of the noise is identified.

Images