CN109241838A - Speed changer quality evaluating method based on psychologic acoustics objective parameter - Google Patents
Speed changer quality evaluating method based on psychologic acoustics objective parameter Download PDFInfo
- Publication number
- CN109241838A CN109241838A CN201810846773.XA CN201810846773A CN109241838A CN 109241838 A CN109241838 A CN 109241838A CN 201810846773 A CN201810846773 A CN 201810846773A CN 109241838 A CN109241838 A CN 109241838A
- Authority
- CN
- China
- Prior art keywords
- signal
- evaluation model
- data set
- labels
- parameters
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 27
- 238000013210 evaluation model Methods 0.000 claims abstract description 77
- 238000012360 testing method Methods 0.000 claims abstract description 43
- 238000012549 training Methods 0.000 claims abstract description 42
- 238000013441 quality evaluation Methods 0.000 claims abstract description 13
- 238000010801 machine learning Methods 0.000 claims abstract description 5
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 4
- 230000005540 biological transmission Effects 0.000 claims description 43
- 230000005284 excitation Effects 0.000 claims description 7
- 210000005069 ears Anatomy 0.000 claims description 6
- 230000008569 process Effects 0.000 claims description 6
- 101000879629 Streptomyces fradiae Subtilisin inhibitor-like protein 5 Proteins 0.000 claims description 5
- 230000006870 function Effects 0.000 claims description 4
- 238000002372 labelling Methods 0.000 claims description 4
- 238000012545 processing Methods 0.000 claims description 4
- 230000013707 sensory perception of sound Effects 0.000 claims description 4
- 230000001133 acceleration Effects 0.000 claims description 3
- XOFYZVNMUHMLCC-ZPOLXVRWSA-N prednisone Chemical compound O=C1C=C[C@]2(C)[C@H]3C(=O)C[C@](C)([C@@](CC4)(O)C(=O)CO)[C@@H]4[C@@H]3CCC2=C1 XOFYZVNMUHMLCC-ZPOLXVRWSA-N 0.000 claims description 3
- 238000001303 quality assessment method Methods 0.000 claims 1
- 230000008901 benefit Effects 0.000 abstract description 3
- 238000011156 evaluation Methods 0.000 abstract 1
- 238000001514 detection method Methods 0.000 description 16
- 238000000605 extraction Methods 0.000 description 6
- 230000003595 spectral effect Effects 0.000 description 6
- 238000004364 calculation method Methods 0.000 description 5
- 238000004458 analytical method Methods 0.000 description 4
- 230000006378 damage Effects 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 230000005236 sound signal Effects 0.000 description 4
- 238000003745 diagnosis Methods 0.000 description 3
- 238000004519 manufacturing process Methods 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 238000013461 design Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 239000002131 composite material Substances 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000007123 defense Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000003203 everyday effect Effects 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000007935 neutral effect Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 238000010183 spectrum analysis Methods 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Theoretical Computer Science (AREA)
- Artificial Intelligence (AREA)
- Signal Processing (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
Abstract
The invention discloses a kind of speed changer quality evaluating methods based on psychologic acoustics objective parameter, signal evaluation model is first built based on machine learning algorithm, signal evaluation model is trained and is tested again, quality evaluation then is carried out to speed changer to be measured using signal evaluation model;The signal evaluation model is trained and is tested using following steps: A, being acquired vibration and noise signals and is marked label;B, signal characteristic is extracted;C, training dataset and test data set are established;D, signal evaluation model is trained;E, signal evaluation model is tested;The recognition accuracy of statistical signal evaluation model, if recognition accuracy repeats step D~E lower than the recognition accuracy of setting, until recognition accuracy is higher than the recognition accuracy of setting.The advantages that present invention has evaluation result objective and accurate, advantageously reduces testing staff's labor intensity, reduces testing cost.
Description
Technical Field
The invention relates to the technical field of transmission detection, in particular to a transmission quality evaluation method based on psychoacoustic objective parameters.
Background
The power train plays an important role as an important assembly of the automobile in transmitting power from the engine to the drive wheels. The transmission is located at the output of the engine and functions to change the transmission ratio, reverse and interrupt power transmission using neutral. The statistical data of the automobile assembly complaint quantity show that the transmission line complaint quantity accounts for 25.88 percent of the total complaint quantity and is ranked the second. Transmission quality issues are directly related to vehicle driveability, which is closely linked to the life safety requirements of occupants. The product with poor quality can not only harm the life health and property loss of passengers, but also damage the brand value and waste the production cost of enterprises. In addition, the transmission is an important component in the power train, and NVH (Noise, Vibration, Harness) problems such as squeal and knocking seriously affect the comfort experience of users. In order to reduce the occurrence of the events, ensure the safety, reliability, economy and comfort of the automobile, a transmission NVH quality evaluation system is researched, the overall design, manufacture and assembly quality of the automobile is guaranteed at the level of the transmission part, and the acceptance of users and markets is improved, so that the last line of defense for economic benefit is improved.
At present, a transmission NVH quality evaluation system researched and developed by an enterprise can detect and compare various analysis indexes of a vibration noise signal of a tested piece in real time, and professional auditors listen and judge the vibration noise signal, so that the quality of the outgoing transmission is qualified on different indexes of vibration and sound. But the transmission offline detection table with personnel participation has a large pain point: firstly, the feeling of a person on sound is subjective, the influence of individual difference is large, and fatigue is caused by long-time work, so that inaccurate judgment or errors exist; secondly, the statistical index of the vibration noise signals detected and analyzed by the offline detection table is too severe due to the requirement of detection precision, and under the abnormal condition of weak fault or composite fault, the detection task is more challenging due to the unstable vibration noise signals with a large amount of noise, so that most of qualified products in the detection table are treated as unqualified products; thirdly, the labor cost is high for enterprises, the participation of people is reduced as much as possible, and the repeated labor of replacing people with machines and programs is imperative. Therefore, in order to meet the requirement of an enterprise on the identification precision of the detection table, reduce the identification error rate and save the cost for the enterprise, it is very important to research the off-line detection table vibration noise signal identification.
The widely adopted NVH fault recognition method for the transmission at present is to analyze various signals such as vibration noise and the like in an expert system supported by an artificial intelligence theory. Although there are many detection indexes in academic results, only a small part is applied to the expert system. This is because the industrial inspection needs to provide the state monitoring statistical indexes aiming at different parts and at different fault levels, which needs to embody the state monitoring indexes.
The transmission is a complex rotating machine, when a fault occurs, a non-stationary signal and a non-Gaussian distribution signal are taken as main signals, and how to extract key fault information from the non-stationary signal becomes the key of fault diagnosis of the transmission. The common fault diagnosis process mainly comprises four parts, namely signal acquisition, feature extraction, feature selection and fusion and fault identification. In the first stage, the signal acquisition mainly adopts modes of vibration, noise, temperature, precise detection and the like. In the second and third stages, whether the signal processing mode is the same and whether the fault mode is the same can cause a large difference of the final index quantity, so that the research on feature extraction becomes significant. In the final stage, fault identification is mainly judged according to extracted characteristic information, previous knowledge, experience and the like. Although the feature extraction method has theoretically achieved abundant results, there are still many problems for industrial mass application: because the feature extraction is directly participated by people, the subjectivity is strong, the workload is large based on statistical analysis, the quality of the feature extraction also directly relates to the accuracy of fault identification, and the judgment based on experience also leads to the increase of learning threshold and more serious judgment errors. With the continuous development of computer hardware level and information processing technology, researchers gradually turn from "design feature" to "feature learning", that is, an intelligent diagnosis technology for researching automatic learning fault features of machines becomes a new research trend. Some researchers also obtain better recognition accuracy in partial fields by combining mature and simple feature extraction technology with the machine learning algorithm which is different day by day. The psychoacoustic parameters are objective physical quantities for describing subjective feeling difference degrees caused by different sound signals, and the psychoacoustic objective parameters are adopted to analyze the sound signals, so that the difference of auditory feeling can be quantitatively analyzed, and the influence of individual subjective feeling is eliminated. However, how to evaluate the quality of the transmission through psychoacoustic objective parameters becomes an urgent problem to be solved.
Disclosure of Invention
Aiming at the defects of the prior art, the technical problems to be solved by the invention are as follows: how to provide a transmission quality evaluation method based on psychoacoustic objective parameters, which has objective and accurate judgment results, is beneficial to reducing the labor intensity of detection personnel and reducing the detection cost.
In order to solve the technical problems, the invention adopts the following technical scheme:
a transmission quality evaluation method based on psychoacoustic objective parameters is characterized in that a signal evaluation model is built based on a machine learning algorithm, then the signal evaluation model is trained and tested, and then the quality of a transmission to be tested is evaluated by the signal evaluation model; the signal evaluation model is trained and tested by adopting the following steps:
A. collecting vibration noise signals and labeling: the method comprises the steps that a transmission is installed on a test rack, vibration noise signals in the running process of the transmission are collected on line in real time under an unsteady state working condition, meanwhile, a auditor conducts test audition, and labels are marked on the collected vibration noise signals according to the test audition result and are respectively qualified labels or unqualified labels, wherein the unqualified labels specifically comprise gear bump labels, input shaft bump labels, shifting fork interference labels, squeaking labels and other unqualified labels;
B. extracting signal characteristics: b, performing data processing on the collected vibration noise signal to obtain psychoacoustic parameters of the vibration noise signal, and marking the label marked on the vibration noise signal in the step A on the corresponding psychoacoustic parameters;
C. establishing a training data set and a testing data set: dividing the psychoacoustic parameters marked with the labels into two groups, and respectively establishing a psychoacoustic parameter training data set and a psychoacoustic parameter testing data set;
D. training a signal evaluation model: inputting a psychoacoustic parameter training data set and corresponding labels into the signal evaluation model, and identifying different psychoacoustic parameters and corresponding labels by a training machine;
E. and (3) testing a signal evaluation model: randomly extracting psychoacoustic parameters from a psychoacoustic parameter test data set, inputting the psychoacoustic parameters into the signal evaluation model, identifying and outputting a label by the signal evaluation model, comparing the output label with a corresponding label of the input psychoacoustic parameters, indicating that the identification is accurate if the output label is consistent with the corresponding label of the input psychoacoustic parameters, and indicating that the identification is wrong if the output label is consistent with the corresponding label of the input psychoacoustic parameters; and D, counting the identification accuracy of the signal evaluation model, and if the identification accuracy is lower than the set identification accuracy, repeating the steps D-E until the identification accuracy is higher than the set identification accuracy.
Further, in the step B, the psychoacoustic parameters include a speech interference level, a signal total loudness and a sharpness, and labels labeled on the vibration noise signal in the step a are labeled on the corresponding speech interference level, the signal total loudness and the sharpness; in the step C, a voice interference level data set, a signal total loudness data set and a sharpness data set are respectively established for the voice interference level, the signal total loudness and the sharpness of the labeled labels; dividing the voice interference level data set, the signal total loudness data set and the sharpness data set into two groups respectively, and establishing a voice interference level training data set, a voice interference level testing data set, a signal total loudness training data set, a signal total loudness testing data set, a sharpness training data set and a sharpness testing data set; in the step D, the voice interference level parameters and the corresponding labels in the voice interference level training data set are input into the signal evaluation model, and a training machine identifies different voice interference level parameters and corresponding labels; inputting a signal total loudness parameter and a corresponding label in a signal total loudness training data set into the signal evaluation model, and identifying different signal total loudness parameters and corresponding labels by a training machine; inputting sharpness parameters and corresponding labels in a sharpness training data set into the signal evaluation model, and identifying different sharpness parameters and corresponding labels by a training machine; in the step E, randomly extracting the voice interference level parameters from the voice interference level test data set, inputting the voice interference level parameters into the signal evaluation model, identifying and outputting the labels by the signal evaluation model, comparing the output labels with the corresponding labels of the input voice interference level parameters, and counting the identification accuracy rate of the voice interference level parameters by the signal evaluation model; randomly extracting a signal total loudness parameter from a signal total loudness test data set, inputting the signal total loudness parameter into the signal evaluation model, identifying and outputting a label by the signal evaluation model, comparing the output label with a corresponding label of the input signal total loudness parameter, and counting the identification accuracy rate of the signal total loudness parameter by the signal evaluation model; randomly extracting sharpness parameters from the sharpness test data set, inputting the sharpness parameters into the signal evaluation model, identifying and outputting labels by the signal evaluation model, comparing the output labels with corresponding labels of the input sharpness parameters, and counting the identification accuracy rate of the sharpness parameters by the signal evaluation model; and D, comparing the recognition accuracy of the signal evaluation model on the speech interference level parameter, the signal total loudness parameter and the sharpness parameter, selecting the psychoacoustic parameter with the highest recognition accuracy as a main recognition object, comparing the recognition accuracy of the signal evaluation model on the psychoacoustic parameter with the set recognition accuracy, and repeating the steps D to E of the psychoacoustic parameter if the recognition accuracy is lower than the set recognition accuracy until the recognition accuracy is higher than the set recognition accuracy.
Further, in the step a, the unsteady state condition refers to an operation condition that the rotation speed of the transmission is accelerated or decelerated at a fixed acceleration.
Further, the speech interference level SIL-5 is calculated using the following formula:
in the formula: l is500Represents an octave value L when the frequency is 500Hz at the center1000,L2000,L4000,L8000Respectively represents the octave values when the central frequencies are 1000Hz, 2000Hz, 4000Hz and 8000 Hz.
Further, the total loudness of the signal is calculated by the following steps:
the characteristic loudness N' (z) is obtained first using the following formula:
in the formula: z: dividing the frequency from 20Hz to 16000Hz which can be heard by human ears into 24 critical frequency bands according to the definition of the Zwicker model theory to obtain each critical band, wherein the unit is Bark;
ETQ: stimuli corresponding to a hearing threshold in a quiet environment;
Eo: reference excitation value, which is the reference sound intensity IO=10-12Tile/meter2Excitation of (2);
e: the calculated excitation corresponding to the sound;
and integrating the characteristic loudness N' (z) at 0-24 Bark to obtain the total loudness:
wherein N' (z) is the specific loudness in sone/Bark; and N is the total loudness.
Further, the sharpness S is calculated by using the following formula:
in the formula: k is a weighting coefficient, and k is 0.11; z is a psychoacoustic critical band in Bark; n' (z) represents the specific loudness in tone/Bark; n is the total loudness; the g (z) function is specifically:
in conclusion, the method has the advantages of objective and accurate judgment result, contribution to reducing the labor intensity of detection personnel, reduction in detection cost and the like.
Drawings
Fig. 1 is a spectrum diagram of a noise signal.
Fig. 2 is an 1/1 octave plot of a noise signal.
FIG. 3 is an octave band diagram.
Fig. 4 is an absolute threshold curve in a Zwicker loudness model.
Fig. 5 to 10 are schematic diagrams of the speech interference level SIL-5.
Fig. 11 is a flowchart of quality evaluation.
Detailed Description
The invention will be described in further detail below with reference to a quality evaluation of a transmission of the MF62 type from a transmission manufacturer.
In order to ensure normal production and operation, the signal evaluation model and the test listening of the listening personnel are synchronously carried out in the whole process, and the following steps are specifically adopted; the method comprises the steps of installing a vibration sensor on a test bench, accelerating the rotating speed of a transmission at a fixed acceleration, collecting vibration signals in the running process of the transmission on line in real time, simultaneously carrying out test listening by a listening person, marking the collected vibration noise signals into labels according to the test listening result, wherein the labels are respectively qualified labels or unqualified labels, and the unqualified labels specifically comprise gear bump damage labels, input shaft bump damage labels, shifting fork interference labels, squealing labels and other unqualified labels.
As shown in fig. 11, in a specific implementation, after a listening person is on duty, the listening person needs a certain time to enter a working state, and after continuous working, the accuracy of a judgment result is affected due to fatigue and the like, so as to acquire more accurate data, transmission data detected from half an hour to half an hour every day after the morning is collected, and at least 500 groups of data are cumulatively sampled.
And carrying out data processing on the collected vibration noise signal to obtain psychoacoustic parameters of the vibration noise signal, wherein the psychoacoustic parameters comprise a voice interference level, a signal total loudness and a sharpness, and labeling labels for labeling the vibration noise signal on the corresponding voice interference level, the signal total loudness and the sharpness.
Wherein, the speech interference level, the signal total loudness and the sharpness are respectively calculated by adopting the following modes:
the speech interference level SIL-5 is calculated using the following formula:
in the formula: l is500Represents an octave value L when the frequency is 500Hz at the center1000,L2000,L4000,L8000Respectively represents the octave values when the central frequencies are 1000Hz, 2000Hz, 4000Hz and 8000 Hz.
Octave: the frequency range of the sound signal which can be heard by human ears is 20Hz to 20KHz, and the spectral analysis of the sound signal generally does not need to carry out specific analysis on each frequency component. For convenience, one divides the audio frequency range of 20Hz to 20KHz into several segments, with each band being a frequency interval. The frequency interval is divided by a constant bandwidth ratio, namely, the ratio of the upper limit and the lower limit of the frequency band is kept to be a constant. The frequency range is determined by the central frequency value, the upper limit frequency value and the lower limit frequency value, and the difference between the upper limit frequency and the lower limit frequency is called the frequency range bandwidth.
Octave center frequency calculation formula:
the center frequencies of the frequency bands are well defined in both ISO and ANSI approved methods. One method is to use 2 as the base, the ratio of two adjacent center frequencies is 2^ (1/N), when N is 3, 1/3 octaves, and other octaves are similar. Another method is to use 10 as the base and the ratio of two adjacent center frequencies is 10^ (3[10N ]), which can also be written as 2^ (3/[10Nlog2 ]). The realistic effect of the two ratios is nearly the same, but if single frequency signals on the band boundaries are of interest, different ratios may cause these signals to appear in different octave bands. Radix 2 is simpler to use, but radix 10 is actually a more reasonable number. In the national standard GB 3240-1982: the usual frequencies in acoustic measurements are the radix-10 method.
The theoretical center frequency and frequency band are calculated based on ISO R266 and ANSI S1.6-1984, and in practical applications, the center frequency is usually used as an approximation thereof, i.e., as a theoretical value. While the actual center frequency calculation is according to International Electrotechnical Commission (IEC) recommendations,
fc=1000×103n/30
wherein,
fcis the actual center frequencyThe value of the rate is,
n represents the number of frequency bands, and n is 0, ± 1, ± 2.
The reference frequency of the octave of the audio part is 1000 Hz. Since the audio aspect of work is generally less concerned with these very low frequency components. Generally, the number of frequency bands is calculated from 1 for convenience, even if sounds having frequencies less than 20Hz are not heard by human ears.
Relation among center frequency value, upper limit frequency value and lower limit frequency value
The division of the frequency interval is expressed by the ratio of the upper and lower limit frequencies, i.e.
Each octave is called by a center frequency, and the relationship between the center frequency and the upper limit frequency, the lower limit frequency and the bandwidth is
In the formula,
fuis an upper limit frequency value
flFor the lower frequency value, the value of the frequency,
flfor the value of the actual center frequency,
Δ f is the bandwidth of the frequency bin,
n is an octave multiple, and if n is 1, it is called 1 octave, and is simply referred to as an octave.
According to the formula, the upper limit frequency value and the lower limit frequency value can be obtained by knowing the center frequency.
The relative width of the frequency ranges is constant (the frequency ranges are divided by a constant bandwidth ratio, namely the ratio of the upper limit and the lower limit of the frequency ranges is kept constant and is 1 octaveThe belt width ratio is 21/2-2-1/20.707, 1/3 octaves bandwidth ratio of) The absolute width increases proportionally with the increase of the center frequency.
Octave calculation of vibration noise data:
fig. 1 and 2 show a spectrogram and 1/1 octave diagram of a noise signal, where fig. 1 is composed of a single spectral line and fig. 2 is composed of corresponding octave bands.
When the octave calculation is carried out, the upper and lower limit frequencies (octave bands) of each octave band are determined according to the corresponding method (base 10 or base 2), so that the number of spectral lines in the corresponding octave band is determined. The sound pressure mean square value in each octave (1/N octave) band is the sum of the mean square values of the spectral line amplitudes in the band:
in the formula, piIs the mean square value of each spectral line,
for the mean square value of sound pressure in each octave (1/N octave) band
Then, the decibel value is calculated according to the formula
SPLband(dB) is the total sound pressure level in dB,
for the mean square value of the sound pressure in each octave (1/N octave) band,
pref=2.0×10-5reference sound pressure, i.e. the sound pressure that the human ear can just hear.
Thus, the octaves represent the sum of the acoustic energy within the corresponding octave band, as shown in FIG. 3. The darker lines indicate the corresponding spectral components, and the denser the corresponding spectral lines, since the higher octave bands are wider. It can also be seen from fig. 3 that when an octave is used for noise testing, the average amplitude of noise within a complete octave is evaluated, and is often used for analyzing broadband noise, and main frequency components are not obvious.
Fig. 5-10 show the results of the SIL-5 calculation analysis with time on the abscissa and sound pressure level in dB on the ordinate. For different types of label signals, the data characteristics are different, and the characteristics are the basis for identifying the machine; in this embodiment, the label of FIG. 5 is pass; the label of fig. 6 is input shaft gouge class; the label of fig. 7 is a gear gouge category; the labels of fig. 8 are of the fork interference type, and the labels of fig. 9 are squealing; the labels of fig. 10 are otherwise rejected. When machine learning is carried out, numerical results obtained by calculating each analysis item are input into the model for learning.
The total loudness of the signal is calculated by adopting the following steps:
loudness is a psychoacoustic parameter that reflects the subjective perception of sound intensity by the human ear, and is expressed in sones (sone). According to the definition of Zwicker theory, the frequency of 20Hz 16000Hz which can be heard by human ears is divided into 24 critical bands (Bark).
The conversion formula between frequency (Hz) and critical band (Bark) is as follows:
Z=13×tan-1(0.00076f)+3.5×tan-1(f/7500)2
the following table shows the relationship between frequency and characteristic frequency band
TABLE 1 frequency-characteristic band relationship correspondence table
The specific loudness is the loudness in each critical frequency band, and the specific loudness N' (z) is obtained by using the following formula:
in the formula:
z: dividing the frequency from 20Hz to 16000Hz which can be heard by human ears into 24 critical frequency bands according to the definition of the Zwicker model theory to obtain each critical band, wherein the unit is Bark;
ETQ: stimuli corresponding to a hearing threshold in a quiet environment;
Eo: reference excitation value, which is the reference sound intensity IO=10-12Tile/meter2Excitation of (2);
e: the calculated excitation corresponding to the sound;
in the formula, the excitation level is generally replaced with a sound pressure level, and a sound pressure level corresponding to the threshold in a quiet environment can be obtained by fig. 4, and thus N' (z) can be obtained.
And integrating the characteristic loudness N' (z) at 0-24 Bark to obtain the total loudness:
where N' (z) is the specific loudness in the Bark domain.
Further, the sharpness is calculated by the following formula:
in the formula: acum is a unit of sharpness S, k is a weighting coefficient, and k is 0.11; z is a psychoacoustic critical band in Bark; n' (z) represents the specific loudness in tone/Bark; n is the total loudness; the g (z) function is specifically:
when Bark number z is more than 16, namely critical bandwidth number is more than 16Bark, the sound sharpness is obviously improved, therefore, a weighting coefficient g (z) is introduced when calculating the sharpness
Respectively establishing a voice interference level data set, a signal total loudness data set and a sharpness data set according to the voice interference level, the signal total loudness and the sharpness of the labeled labels; and dividing the voice interference level data set, the signal total loudness data set and the sharpness data set into two groups respectively, establishing a voice interference level training data set, a voice interference level testing data set, a signal total loudness training data set, a signal total loudness testing data set, a sharpness training data set and a sharpness testing data set, ensuring that the voice interference level training data set, the signal total loudness training data set and the sharpness training data set contain 400 data, and ensuring that the voice interference level testing data set, the signal total loudness testing data set and the sharpness testing data set contain 100 data.
Inputting the voice interference level parameters and the corresponding labels in the voice interference level training data set into the signal evaluation model, and training a machine to identify different voice interference level parameters and corresponding labels; inputting a signal total loudness parameter and a corresponding label in a signal total loudness training data set into the signal evaluation model, and identifying different signal total loudness parameters and corresponding labels by a training machine; sharpness parameters and corresponding labels in a sharpness training data set are input into the signal evaluation model, and a training machine identifies different sharpness parameters and corresponding labels.
Randomly extracting voice interference level parameters from a voice interference level test data set, inputting the voice interference level parameters into the signal evaluation model, identifying and outputting a label by the signal evaluation model, comparing the output label with a corresponding label of the input voice interference level parameters, and counting the identification accuracy rate of the voice interference level parameters by the signal evaluation model; randomly extracting a signal total loudness parameter from a signal total loudness test data set, inputting the signal total loudness parameter into the signal evaluation model, identifying and outputting a label by the signal evaluation model, comparing the output label with a corresponding label of the input signal total loudness parameter, and counting the identification accuracy rate of the signal total loudness parameter by the signal evaluation model; randomly extracting sharpness parameters from the sharpness test data set, inputting the sharpness parameters into the signal evaluation model, identifying and outputting labels by the signal evaluation model, comparing the output labels with corresponding labels of the input sharpness parameters, and counting the identification accuracy rate of the sharpness parameters by the signal evaluation model.
Through comparison, the recognition accuracy of the signal evaluation model on the voice interference level parameters is relatively high, the voice interference level parameters are selected as main recognition objects, data in a voice interference level training data set are added, and the signal evaluation model is trained until the recognition accuracy of the signal evaluation model on the voice interference level parameters reaches 90% of the set recognition accuracy.
The off-line transmission is tested and identified by the trained signal evaluation model and the auditors simultaneously and respectively, the identification result of the signal evaluation model is compared with the identification result of the auditors, and in a factory, in order to ensure that the audition result is more accurate, three professional auditors are generally adopted for audition, and a conclusion is comprehensively obtained. Through the detection of the consecutive day, the transmission 110 stations are detected in total, and the detected transmissions are numbered in the detection order.
The coincidence rate of the recognition result of the signal evaluation model and the recognition result of the listening personnel is 78%, the coincidence rate of the recognition result of the signal evaluation model and the recognition result of the listening personnel is increased to 80% by rechecking the inconsistent transmission, the recognition result of the signal evaluation model and the recognition result of the previous time are basically unchanged in the rechecking process, the increase of the rechecking coincidence rate is mainly used for the listening personnel to correct part of the recognition results, wherein the transmission numbers for correcting the recognition results are mainly concentrated at the beginning stage and the ending stage of the morning and afternoon work, which indicates that the listening results of professional listening engineers are greatly influenced by fatigue and are easy to misjudge. In addition, when more than two unqualified types exist, the listening personnel can only make unqualified judgment and cannot accurately locate the specific unqualified type.
The above description is only exemplary of the present invention and should not be taken as limiting, and any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (6)
1. A transmission quality evaluation method based on psychoacoustic objective parameters is characterized in that a signal evaluation model is built based on a machine learning algorithm, then the signal evaluation model is trained and tested, and then the quality of a transmission to be tested is evaluated by the signal evaluation model; the signal evaluation model is trained and tested by adopting the following steps:
A. collecting vibration noise signals and labeling: the method comprises the steps that a transmission is installed on a test rack, vibration noise signals in the running process of the transmission are collected on line in real time under an unsteady state working condition, meanwhile, a auditor conducts test audition, and labels are marked on the collected vibration noise signals according to the test audition result and are respectively qualified labels or unqualified labels, wherein the unqualified labels specifically comprise gear bump labels, input shaft bump labels, shifting fork interference labels, squeaking labels and other unqualified labels;
B. extracting signal characteristics: b, performing data processing on the collected vibration noise signal to obtain psychoacoustic parameters of the vibration noise signal, and marking the label marked on the vibration noise signal in the step A on the corresponding psychoacoustic parameters;
C. establishing a training data set and a testing data set; dividing the psychoacoustic parameters marked with the labels into two groups, and respectively establishing a psychoacoustic parameter training data set and a psychoacoustic parameter testing data set;
D. training a signal evaluation model: inputting a psychoacoustic parameter training data set and corresponding labels into the signal evaluation model, and identifying different psychoacoustic parameters and corresponding labels by the training evaluation model;
E. and (3) testing a signal evaluation model: inputting a psychoacoustic parameter test data set into the signal evaluation model, identifying and outputting a label by the signal evaluation model, comparing the output label with a corresponding label of the input psychoacoustic parameter, if the output label is consistent with the corresponding label of the input psychoacoustic parameter, indicating that the identification is accurate, otherwise, indicating that the identification is wrong; and D, counting the identification accuracy of the signal evaluation model, and if the identification accuracy is lower than the set identification accuracy, repeating the steps D-E until the identification accuracy is higher than the set identification accuracy.
2. The transmission quality evaluation method based on psychoacoustic objective parameters according to claim 1, wherein in said step B, said psychoacoustic parameters include a speech disturbance level, a signal total loudness and a sharpness, and labels labeled for vibration noise signals in step a are labeled on the corresponding speech disturbance level, signal total loudness and sharpness; in the step C, a voice interference level data set, a signal total loudness data set and a sharpness data set are respectively established for the voice interference level, the signal total loudness and the sharpness of the labeled labels; dividing the voice interference level data set, the signal total loudness data set and the sharpness data set into two groups respectively, and establishing a voice interference level training data set, a voice interference level testing data set, a signal total loudness training data set, a signal total loudness testing data set, a sharpness training data set and a sharpness testing data set; in the step D, the voice interference level parameters and the corresponding labels in the voice interference level training data set are input into the signal evaluation model, and a training machine identifies different voice interference level parameters and corresponding labels; inputting a signal total loudness parameter and a corresponding label in a signal total loudness training data set into the signal evaluation model, and identifying different signal total loudness parameters and corresponding labels by a training machine; inputting sharpness parameters and corresponding labels in a sharpness training data set into the signal evaluation model, and identifying different sharpness parameters and corresponding labels by a training machine; in the step E, randomly extracting the voice interference level parameters from the voice interference level test data set, inputting the voice interference level parameters into the signal evaluation model, identifying and outputting the labels by the signal evaluation model, comparing the output labels with the corresponding labels of the input voice interference level parameters, and counting the identification accuracy rate of the voice interference level parameters by the signal evaluation model; randomly extracting a signal total loudness parameter from a signal total loudness test data set, inputting the signal total loudness parameter into the signal evaluation model, identifying and outputting a label by the signal evaluation model, comparing the output label with a corresponding label of the input signal total loudness parameter, and counting the identification accuracy rate of the signal total loudness parameter by the signal evaluation model; randomly extracting sharpness parameters from the sharpness test data set, inputting the sharpness parameters into the signal evaluation model, identifying and outputting labels by the signal evaluation model, comparing the output labels with corresponding labels of the input sharpness parameters, and counting the identification accuracy rate of the sharpness parameters by the signal evaluation model; and D, comparing the recognition accuracy of the signal evaluation model on the speech interference level parameter, the signal total loudness parameter and the sharpness parameter, selecting the psychoacoustic parameter with the highest recognition accuracy as a main recognition object, comparing the recognition accuracy of the signal evaluation model on the psychoacoustic parameter with the set recognition accuracy, and repeating the steps D to E of the psychoacoustic parameter if the recognition accuracy is lower than the set recognition accuracy until the recognition accuracy is higher than the set recognition accuracy.
3. The method according to claim 1, wherein the unsteady state condition in step a is an operation condition in which the torque of the transmission is fixed to a non-zero value and the rotational speed is accelerated or decelerated at a fixed acceleration.
4. The transmission quality evaluation method based on psychoacoustic objective parameters according to claim 1, wherein the speech interference level SIL-5 is calculated using the following formula:
in the formula: l is500Represents an octave value L when the frequency is 500Hz at the center1000,L2000,L4000,L8000Respectively represents the octave values when the central frequencies are 1000Hz, 2000Hz, 4000Hz and 8000 Hz.
5. The transmission quality assessment method based on psychoacoustic objective parameters according to claim 1, wherein the total loudness of the signal is calculated by the following steps:
the characteristic loudness N' (z) is obtained first using the following formula:
in the formula: z: dividing the frequency from 20Hz to 16000Hz which can be heard by human ears into 24 critical frequency bands according to the definition of the Zwicker model theory to obtain each critical band, wherein the unit is Bark;
ETQ: stimuli corresponding to a hearing threshold in a quiet environment;
Eo: reference excitation value, which is the reference sound intensity IO=10-12Tile/meter2Excitation of (2);
e: the calculated excitation corresponding to the sound;
and integrating the characteristic loudness N' (z) at 0-24 Bark to obtain the total loudness:
wherein N' (z) is the specific loudness in sone/Bark; and N is the total loudness.
6. A transmission quality evaluation method based on psychoacoustic objective parameters according to claim 1, characterized in that said sharpness S is calculated using the following formula:
in the formula: k is a weighting coefficient, and k is 0.11; z is a psychoacoustic critical band in Bark; n' (z) represents the specific loudness in tone/Bark; n is the total loudness; the g (z) function is specifically:
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810846773.XA CN109241838A (en) | 2018-07-27 | 2018-07-27 | Speed changer quality evaluating method based on psychologic acoustics objective parameter |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810846773.XA CN109241838A (en) | 2018-07-27 | 2018-07-27 | Speed changer quality evaluating method based on psychologic acoustics objective parameter |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109241838A true CN109241838A (en) | 2019-01-18 |
Family
ID=65073176
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810846773.XA Pending CN109241838A (en) | 2018-07-27 | 2018-07-27 | Speed changer quality evaluating method based on psychologic acoustics objective parameter |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109241838A (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110261101A (en) * | 2019-06-14 | 2019-09-20 | 西安交通大学 | Quantitative evaluation method of uttering long and high-pitched sounds based on orthogonal-pairs of comparative test and SVR |
CN112506341A (en) * | 2020-12-01 | 2021-03-16 | 瑞声新能源发展(常州)有限公司科教城分公司 | Vibration effect generation method and device, terminal equipment and storage medium |
CN112504673A (en) * | 2020-12-07 | 2021-03-16 | 中国科学院重庆绿色智能技术研究院 | Carrier roller fault diagnosis method and system based on machine learning and storage medium |
CN114383711A (en) * | 2020-10-06 | 2022-04-22 | 丰田自动车株式会社 | Abnormal sound determination device for vehicle |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101672690A (en) * | 2009-09-27 | 2010-03-17 | 吉林大学 | Method for objectively and quantifiably evaluating noise fret degree in vehicle based on auditory model |
CN102539151A (en) * | 2012-01-18 | 2012-07-04 | 北京工业大学 | Intelligent online quality detection method for automobile transmission |
CN103323532A (en) * | 2012-03-21 | 2013-09-25 | 中国科学院声学研究所 | Fish identification method and system based on psychoacoustics parameters |
CN103471709A (en) * | 2013-09-17 | 2013-12-25 | 吉林大学 | Method for predicting noise quality of noise inside passenger vehicle |
CN105021275A (en) * | 2015-07-15 | 2015-11-04 | 吉林大学 | Time domain dynamic characteristic analysis-based intra-vehicle sound quality prediction method |
-
2018
- 2018-07-27 CN CN201810846773.XA patent/CN109241838A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101672690A (en) * | 2009-09-27 | 2010-03-17 | 吉林大学 | Method for objectively and quantifiably evaluating noise fret degree in vehicle based on auditory model |
CN102539151A (en) * | 2012-01-18 | 2012-07-04 | 北京工业大学 | Intelligent online quality detection method for automobile transmission |
CN103323532A (en) * | 2012-03-21 | 2013-09-25 | 中国科学院声学研究所 | Fish identification method and system based on psychoacoustics parameters |
CN103471709A (en) * | 2013-09-17 | 2013-12-25 | 吉林大学 | Method for predicting noise quality of noise inside passenger vehicle |
CN105021275A (en) * | 2015-07-15 | 2015-11-04 | 吉林大学 | Time domain dynamic characteristic analysis-based intra-vehicle sound quality prediction method |
Non-Patent Citations (4)
Title |
---|
《计量测试技术手册》编辑委员会: "《计量测试技术手册 第9卷 声学》", 30 June 1997, 中国计量出版社 * |
吴小珊: "变速器声品质室内评价方法研究", 《中国优秀硕士学位论文全文数据库工程科技Ⅱ辑》 * |
应怀樵: "《现代振动与噪声技术》", 30 April 2017, 航空工业出版社 * |
施全 等: "变速器啸叫声品质的RBF神经网络预测与权重分析", 《振动与冲击》 * |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110261101A (en) * | 2019-06-14 | 2019-09-20 | 西安交通大学 | Quantitative evaluation method of uttering long and high-pitched sounds based on orthogonal-pairs of comparative test and SVR |
CN110261101B (en) * | 2019-06-14 | 2021-07-27 | 西安交通大学 | Howling quantitative evaluation method based on orthogonal-pair comparison test and SVR |
CN114383711A (en) * | 2020-10-06 | 2022-04-22 | 丰田自动车株式会社 | Abnormal sound determination device for vehicle |
CN114383711B (en) * | 2020-10-06 | 2024-03-15 | 丰田自动车株式会社 | Abnormal sound judging device for vehicle |
CN112506341A (en) * | 2020-12-01 | 2021-03-16 | 瑞声新能源发展(常州)有限公司科教城分公司 | Vibration effect generation method and device, terminal equipment and storage medium |
CN112506341B (en) * | 2020-12-01 | 2022-05-03 | 瑞声新能源发展(常州)有限公司科教城分公司 | Vibration effect generation method and device, terminal equipment and storage medium |
CN112504673A (en) * | 2020-12-07 | 2021-03-16 | 中国科学院重庆绿色智能技术研究院 | Carrier roller fault diagnosis method and system based on machine learning and storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109241838A (en) | Speed changer quality evaluating method based on psychologic acoustics objective parameter | |
CN106092310A (en) | A kind of automotive transmission vibration noise off-line test method | |
CN108663115A (en) | A kind of car inside idle noise objective quantification evaluation method | |
CN110261101B (en) | Howling quantitative evaluation method based on orthogonal-pair comparison test and SVR | |
CN108535012B (en) | A kind of transmission squawk NVH performance off-line test method | |
US20060224367A1 (en) | Inspection apparatus, aid device for creating judgement model therefor, abnormality detection device for endurance test apparatus and endurance test method | |
CN112052712B (en) | Power equipment state monitoring and fault identification method and system | |
CN208520476U (en) | A kind of high-speed transplanter noise qualities evaluation system | |
Kannan et al. | Demodulation band optimization in envelope analysis for fault diagnosis of rolling element bearings using a real-coded genetic algorithm | |
CN113899556A (en) | Method for identifying engine gear squeaking noise | |
CN112098939B (en) | Method and device for identifying and evaluating noise pollution source | |
CN112487865A (en) | Automatic loudspeaker classification method based on machine learning | |
CN115165079A (en) | Equipment alarm method based on multiple sound characteristic analysis | |
CN110096733A (en) | A kind of power transmission and transformation noise rating method for establishing model and power transmission and transformation noise rating method | |
CN109781245B (en) | Objective evaluation method for pulse noise of diesel engine | |
CN113343384B (en) | Sound quality subjective and objective evaluation method under variable rotating speed working condition of transmission | |
CN106725289B (en) | A kind of vehicle sound quality evaluation method based on pupil diameter | |
CN116933620A (en) | Noise quality evaluation and model building method | |
CN108344503A (en) | A kind of high-speed transplanter noise qualities evaluation system | |
CN110296847B (en) | Noise testing method for new energy passenger car | |
WO2023060542A1 (en) | Bearing fault detection method and storage medium | |
CN111442936A (en) | Method for confirming sound quality detection limit value and method for confirming sound quality detection fault | |
Sottek | Sound quality evaluation of noises with spectro-temporal patterns | |
Manhertz et al. | Managing measured vibration data for malfunction detection of an assembled mechanical coupling | |
CN117559937B (en) | Impedance matching network adjusting method |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190118 |