In-situ quantitative evaluation method and device for health state of array MEMS microphone in noise environment

Abstract

The invention relates to the field of microphone anomaly detection, in particular to an array MEMS microphone health state in-situ quantitative evaluation method and device in a noise environment. The invention provides a noise environment-oriented in-situ quantitative evaluation method for the health state of an array microphone, and designs an embedded broadband ultrasonic frequency sweeper capable of outputting relatively flat sound wave power at the frequency band of 20k-80kHz. A dual-mode self-checking process design of rapid self-checking and full-channel cross self-checking is adopted, and the 20k-80kHz frequency response characteristic of the microphone can be rapidly detected in a conventional indoor / outdoor noise environment. A Welch method is used for estimating power spectral density, a health state quantitative index is constructed by using PSD correlation coefficients among channels, and screening, marking and removing of abnormal channels are realized by setting a multi-level threshold value, and visual presentation is realized in a graphic and text mode.

Description

Technical Field

[0001] The present invention relates to the field of microphone anomaly detection, and in particular to an in-situ quantitative evaluation method and device for the health status of an array MEMS microphone in a noisy environment. Background Art

[0002] In industrial applications, phenomena such as gas leaks, partial discharges in power equipment, and abnormal mechanical vibration generate ultrasonic waves with a wide frequency band, typically between 20kHz and 100kHz. To avoid interference from ambient audible noise and taking into account the frequency response range of MEMS microphones, sound source identification and location equipment typically selects the 20kHz to 80kHz ultrasonic frequency band for analysis and processing.

[0003] As an advanced sound source identification and location device, acoustic cameras utilize a low-cost MEMS microphone array to collect acoustic signals and apply high-resolution beamforming algorithms to generate visual acoustic images, thereby enabling the identification and location of sound sources such as leaks, discharges, and abnormal noises. Currently, acoustic cameras are widely used in areas such as gas leaks, partial discharges, and vibration noises.

[0004] Acoustic camera arrays typically integrate dozens to hundreds of MEMS microphone channels, with large arrays even having thousands of channels. Existing sound source localization algorithms assume that all microphone channels are functioning normally. However, due to long-term use and environmental factors (such as high dust, high temperature, high humidity, and high salt content), individual microphones in the array will inevitably experience performance degradation or failure, primarily manifesting as abnormal gain and phase frequency response characteristics. If these abnormal channels are not promptly identified and eliminated, they will significantly affect the overall sensitivity and positioning accuracy of the device, necessitating regular assessment of the health of the MEMS microphones.

[0005] Common methods for microphone anomaly detection include amplitude domain analysis, Fourier transform and its inverse transform, and correlation analysis. Correlation analysis measures the relationship between signal variations and the degree of correlation based on the correlation coefficient. It is highly applicable and easy to use, and has been widely used in anomaly detection and fault diagnosis.

[0006] Existing microphone anomaly detection: The applicable frequency is low, ≤35kHz, and it cannot perform wideband scanning, and cannot evaluate the health status of the microphone in the high-frequency ultrasonic band. It also has high requirements for the detection environment and needs to be performed in a quiet environment, even in an anechoic chamber.

[0007] Once hundreds or even thousands of MEMS microphones are assembled, they are difficult to disassemble for inspection. Therefore, a simple and efficient in-situ quantitative evaluation system is urgently needed. This system, which does not require an anechoic chamber or a particularly quiet location, can quickly test the microphone's 20k–80kHz frequency response characteristics in a conventional indoor or outdoor noise environment (background noise is approximately 40–65dB (A-weighted)). It automatically removes degraded and failed microphones from subsequent calculations. It also automatically identifies the location, number, proportion, and channel number of degraded and failed channels, presenting them visually with both graphics and text. When the total proportion of degraded and failed channels exceeds 15% (the repair threshold varies slightly for different arrays), it alerts the user to promptly perform repairs to prevent further failures. Summary of the Invention

[0008] In order to solve the problems of the prior art, the present invention provides a method and device for in-situ quantitative evaluation of the health status of an array MEMS microphone in a noisy environment.

[0009] In order to achieve the above object, the present invention adopts the following technical solutions:

[0010] A method for in-situ quantitative evaluation of the health status of a MEMS microphone array in a noisy environment comprises the following steps:

[0011] The acoustic camera enters the self-test mode and controls the ultrasonic frequency scanner through communication to perform automatic frequency scanning.

[0012] Each microphone in the microphone array responds to the swept frequency sound wave simultaneously. The FPGA collects the waveform and packages the data before transmitting it to the computing unit of the acoustic camera.

[0013] Calculate the Welch power spectral density of all channels and intercept the power spectral density between 20k-80kHz; replace the reference channel Ref_ch in turn, calculate the average value of the cross-correlation coefficient RRef_ch_avg between it and other channels, and determine:

[0014] If the current channel R Ref_ch_avg > Excellent threshold, indicating that this channel is basically normal and can be used as a reference channel, the device will enter the fast self-test mode, which takes about 2 minutes to obtain statistical results and charts; if all channels R Ref_ch_avg If all are ≤ the excellent threshold, it indicates that no qualified reference channel Ref_ch is found, and the device will enter the full-channel cross-check mode. It takes about 2 hours to obtain the statistical results and charts.

[0015] It prompts whether the equipment can be used normally, whether it needs maintenance, and ends the self-test.

[0016] The in-situ quantitative evaluation method for the health status of a MEMS microphone array in a noisy environment, in a rapid self-test mode, comprises the following steps:

[0017] S1) Initialize the reference channel Ref_ch = 1, and calculate the power spectrum density PSD of the Ref_ch channel according to formula (2) 20k–80k Power spectral density PSD of all channels 20k–80k The result is a one-dimensional matrix with the matrix elements rounded to two decimal places. Then the correlation coefficient of the reference channel Ref_ch to itself is assigned to "null". The final result is a one-dimensional matrix.

[0018] S2) Calculate the one-dimensional matrix R Ref_ch The average value of all elements of R Ref_ch_avg , make the following judgment: If R Ref_ch_avg ≤ the excellent threshold, indicating that the current channel is not suitable as a reference channel and needs to be replaced with the next channel, Ref_ch+1, and then repeat step S1);

[0019] If we traverse all channels R Ref_ch_avg If all the values ​​are less than or equal to the excellent threshold, a prompt message will be displayed on the screen of the acoustic camera, and the system will switch to the full-channel cross-check mode.

[0020] If R Ref_ch_avg >Excellent threshold, indicating that the channel is basically normal and can be used as a reference channel; usually a qualified reference channel is quickly found, and the process goes to step S3;

[0021] S3) According to the classification interval and threshold of health degree, the correlation coefficient one-dimensional matrix R is calculated respectively Ref_ch The number, proportion, channel number, and position coordinates of the elements in each interval;

[0022] S4) Processing and presentation of statistical results.

[0023] The in-situ quantitative evaluation method for the health status of a MEMS microphone array in a noisy environment, in a full-channel cross-check mode, comprises the following steps:

[0024] Step 1) Calculate the power spectral density PSD of all channels according to formula (2) 20k–80k Cross-over all channel power spectral density PSD 20k–80k The result is a two-dimensional matrix, and the matrix elements are rounded to two decimal places; then the elements on the main diagonal are assigned to "empty"; the final result is a two-dimensional matrix; then the average value of each row is calculated, rounded to two decimal places, and a new one-dimensional matrix is ​​formed; step 2) according to the classification interval and threshold of healthiness, the correlation coefficient one-dimensional matrix R is statistically calculated respectively. All_ch_avg The number, proportion, channel number, and position coordinates of the elements in each interval;

[0025] Step 3) How to process and present the statistical results.

[0026] The in-situ quantitative assessment method for the health status of a MEMS microphone array in a noisy environment divides a signal X(j) of length N into K segments, each segment of length L, with adjacent segments overlapping by D, typically D = L / 2.

[0027] N=L+(K-1)(LD), and the solution is

[0028] Apply the Hanning window function W(j) to each signal segment and reduce spectrum leakage

[0029]

[0030] Calculate the discrete Fourier transform (DFT) of each windowed segment to convert the time domain signal to the frequency domain

[0031] in Get the modified periodogram of segment K

[0032] The frequency is discretized into f s is the sampling rate, n=0,1,…,L / 2

[0033] The normalization factor U is used to compensate for the energy loss caused by windowing

[0034] All K periodograms are averaged to obtain the final Welch power spectral density estimate

[0035]

[0036] The combined formula is The in-situ quantitative evaluation method for the health status of an array MEMS microphone in a noisy environment is based on the spectrum estimation method, and the autopower spectrum of the two microphones is obtained by formula (1). Calculate the correlation coefficient of two channels

[0037] Among them, cov represents and The covariance between They are The method for in-situ quantitative evaluation of the health status of an array MEMS microphone in a noisy environment involves a device comprising an embedded ultrasonic frequency sweeper, including:

[0038] The power conversion module is a DC / DC power conversion module; the power supply and communication module is connected to the power conversion module;

[0039] The communication control and waveform generation MCU is connected to the power conversion module, and the power supply and communication module is connected to the communication control and waveform generation MCU;

[0040] The integrated amplifier chip is connected to the communication control and waveform generation MCU, and the integrated amplifier chip is connected to the power conversion module;

[0041] Ultrasonic speaker, connected to the integrated amplifier chip.

[0042] Compared with the existing technology, the beneficial effects of the invention are: proposing an in-situ quantitative assessment method for the health status of array microphones in noisy environments, and designing an embedded wideband ultrasonic frequency sweeper that can output relatively flat sound wave power in the 20k-80kHz frequency band. Using a dual-mode self-test process design of "quick self-test" and "full-channel cross-self-test", the 20k-80kHz frequency response characteristics of the microphone can be quickly tested in a conventional indoor / outdoor noise environment (background sound is about 40-65dB (A-weighted)). The Welch method is used to estimate the power spectral density (PSD) and the PSD correlation coefficient between each channel is used to construct a quantitative indicator of health status. By setting multi-level thresholds (qualified, decayed, degraded, and failed), abnormal channels can be screened, marked, and eliminated, and presented intuitively in a graphic and textual manner. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Other features, objects and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings.

[0044] Figure 1 1 is a schematic diagram of microphone signal segmentation according to the present invention.

[0045] Figure 2 This is a principle block diagram of the broadband ultrasonic frequency sweeper module of the present invention.

[0046] Figure 3 This is a circuit schematic diagram of the broadband ultrasonic frequency sweeper module of the present invention.

[0047] Figure 4 It is the external dimension diagram of the ultrasonic speaker of the present invention.

[0048] Figure 5 It is the original frequency response curve of the ultrasonic speaker of the present invention.

[0049] Figure 6 It is a frequency response curve diagram of the ultrasonic speaker after automatic compensation of the excitation voltage of the present invention.

[0050] Figure 7 This is a schematic diagram of the acoustic camera structure of the present invention. Figure 1 .

[0051] Figure 8This is a schematic diagram of the acoustic camera structure of the present invention. Figure 2 .

[0052] Figure 9 This is a schematic diagram of the acoustic camera structure of the present invention. Figure 3 .

[0053] Figure 10 It is a processing flow chart of the present invention.

[0054] Figure 11 1 is a diagram of the array of 120 microphone channels of the acoustic camera of the present invention.

[0055] Figure 12 3 is a power spectrum density diagram of the reference channel Ref_ch according to the first embodiment of the present invention.

[0056] Figure 13 1 is a power spectrum density diagram of all channels in Example 1 of the present invention.

[0057] Figure 14 This is a statistical chart of the results of the fast self-test mode of Example 1 of the present invention.

[0058] Figure 15 This is a coordinate position diagram of the result of the quick self-test mode of Example 1 of the present invention.

[0059] Figure 16 The result statistics of the full channel cross self-test mode of embodiment 1 of the present invention are Figure 1 .

[0060] Figure 17 The result statistics of the full channel cross self-test mode of embodiment 1 of the present invention are Figure 2 .

[0061] Figure 18 This is a result coordinate position diagram of the full-channel cross self-test mode of Example 1 of the present invention.

[0062] Figure 19 : is an array diagram of 60 microphone channels of the acoustic camera of the present invention.

[0063] Figure 20 is the power spectral density of the reference channel Ref_ch in embodiment 2 of the present invention.

[0064] Figure 21 is a power spectrum density diagram of all channels in Example 2 of the present invention.

[0065] Figure 22 This is a statistical chart of the results of the fast self-test mode of Example 2 of the present invention.

[0066] Figure 23 This is a coordinate position diagram of the fast self-test mode of Example 2 of the present invention.

[0067] Figure 24 The result statistics of the full channel cross self-test mode of embodiment 2 of the present invention are Figure 1 .

[0068] Figure 25 The result statistics of the full channel cross self-test mode of embodiment 2 of the present invention are Figure 2 .

[0069] Figure 26 This is a result coordinate position diagram of the full-channel cross self-check mode of Example 2 of the present invention. DETAILED DESCRIPTION

[0070] The present invention is further described in detail below by way of examples. The examples are only used to illustrate the present invention and are not intended to limit the scope of the present invention.

[0071] This solution offers a simple and efficient in-situ quantitative assessment method (without removing microphones from the array). It can quickly detect degraded and failed microphones in typical indoor and outdoor noise environments (background noise levels of approximately 40–65dB (A-weighted)), eliminating them from subsequent calculations. It automatically identifies the location, number, ratio, and channel number of degraded and failed channels, presenting them visually with both graphics and text. A dual-mode self-test process is designed: the "Quick Self-Test" mode automatically searches for qualified reference channels and calculates a one-dimensional correlation coefficient matrix between the reference channel and all channels. This generates statistical results and graphs in approximately two minutes, completing a preliminary screening of abnormal channels and meeting daily O&M requirements. The "Full Channel Cross-Section Self-Test" mode first constructs a two-dimensional correlation coefficient matrix and then analyzes the average values ​​of each row to improve detection robustness. This results in more accurate results, but takes slightly longer, approximately two hours, to generate statistical results and graphs. This makes it suitable for equipment production and maintenance self-tests, and the frequency sweep range can be expanded from 20kHz–80kHz to 1kHz–100kHz.

[0072] In the specific implementation, this solution needs to design an embedded broadband ultrasonic frequency sweeper (module), focusing on the 52k-80kHz frequency band where the speaker has a poor frequency response ( Figure 5 ) excitation voltage is automatically compensated, so that the speaker can output a relatively flat sound wave power curve in the 20k-80kHz frequency band ( Figure 6 ).

[0073] The module circuit obtains a DC12V / 1A operating power supply from the acoustic camera and interacts with the acoustic camera via RS232 serial communication. Using an STM32F103 microcontroller chip, it outputs a small sinusoidal signal with an adjustable frequency of 1k–100kHz and an adjustable amplitude of 0–3.3V. This signal is then driven by an LM386 integrated amplifier and a piezoelectric crystal speaker. It can output ultrasonic waves at any fixed frequency between 1k–100kHz, or automatically sweep the frequency using a combination of "frequency modulation steps of 10k / 1k / 100 / 10 / 1Hz" and "time steps of 1 / 3 / 5 / 10s." The present invention is programmed to automatically sweep the frequency at "20k–80kHz @ frequency modulation steps of 1kHz @ time steps of 1s," completing a rapid sweep cycle in 60 seconds.

[0074] Embedded ultrasonic scanner (module) includes:

[0075] The power conversion module 1 is a DC / DC power conversion module; the power supply and communication module 6 is connected to the power conversion module 1;

[0076] The communication control and waveform generation MCU 2 is connected to the power conversion module 1, and the power supply and communication module 6 is connected to the communication control and waveform generation MCU 2;

[0077] The integrated amplifier chip 4 is connected to the communication control and waveform generation MCU 2, and the integrated amplifier chip 4 is connected to the power conversion module 1;

[0078] Ultrasonic speaker 5, connected to the integrated amplifier chip 4;

[0079] in:

[0080] ① Power supply and communication module 6: The J1 socket obtains DC12V / 1A working power from the acoustic camera and performs RS232 serial communication with the acoustic camera to exchange control information.

[0081] ②Power conversion module 1: +12V power passes through D6 / M7 (anti-reverse) diode, inputs U5 / AMS1117 voltage regulator chip and outputs DC3.3V / 1A working voltage.

[0082] ③ Communication Control and Waveform Generation MCU2: Utilizes the U2 / STM32F103 microcontroller chip, enabling RS232 serial communication with an external host computer via the J1 socket. It starts or stops waveform signal output, and indicates operating status via the D5 / LED flashing frequency.

[0083] ④ Program-controlled sine wave small signal 3: The DAC is triggered by the timer of the U2 / STM32F103 microcontroller chip.

[0084] The built-in sine waveform data table is converted into an analog voltage, outputting a small sine wave signal with an adjustable frequency from 1k–100kHz and an adjustable amplitude from 0–3.3V. Based on the pre-measured and preset speaker frequency response curve, the amplitude of the small sine wave signal is programmable by frequency band, automatically compensating for the excitation voltage in the 52k–80kHz frequency band, where the ultrasonic speaker's frequency response is poor.

[0085] ⑤ Integrated amplifier chip 4: This uses the U3 / LM386 integrated amplifier chip, with an operating voltage of 12V and a frequency response range of 40Hz–100kHz. It amplifies the small sine wave signal sent from U2 / STM32F103 and outputs it to the ultrasonic speaker through the J2 connector.

[0086] ⑥ Ultrasonic speaker 5: There is no suitable broadband ultrasonic speaker at present. After comparative testing, a small, economical and readily available circular piezoelectric crystal speaker ( Figure 4 ). It is necessary to use the 52k–80kHz band with poor frequency response ( Figure 5 ) excitation voltage is automatically compensated so that it can output a relatively flat sound wave power curve in the 20k–80kHz frequency band ( Figure 6 ).in Figure 5 Medium, the frequency response above 52kHz drops off sharply / is poor; Figure 6 The frequency response from 10k to 80kHz is relatively flat / good.

[0087] In order to facilitate portability and self-inspection at any time, an ultrasonic scanner (module) needs to be embedded in the acoustic camera;

[0088] in:

[0089] Form 1: The embedded ultrasonic frequency scanner 72 and ultrasonic speaker 73 are built into the acoustic camera 7 and can be used at any time.

[0090] Form 2: An external embedded ultrasonic frequency sweeper 75 is provided on one side of the acoustic camera 7, and the ultrasonic frequency sweeper 75 is connected to the acoustic camera via a connector 76. The ultrasonic frequency sweeper 75 is connected to an ultrasonic speaker 74, and the ultrasonic speaker 74 faces the microphone array 711 (of the acoustic camera); the ultrasonic speaker 74 and the ultrasonic frequency sweeper 75 are external, and when in use, they are plugged into the acoustic camera 7 via a connector 76 for use.

[0091] Either form 1 or form 2 can be used.

[0092] Wherein: 712 is the display screen of the acoustic camera, 711 is the microphone array of the acoustic camera;

[0093] Automated testing process:

[0094] A method for in-situ quantitative evaluation of the health status of a MEMS microphone array in a noisy environment, characterized by comprising the following steps:

[0095] The acoustic camera enters the self-test mode and controls the ultrasonic frequency scanner through communication to perform automatic frequency scanning.

[0096] Each microphone in the microphone array responds to the swept frequency sound wave simultaneously. The FPGA collects the waveform and packages the data before transmitting it to the computing unit of the acoustic camera.

[0097] Calculate the Welch power spectral density of all channels and intercept the power spectral density between 20k-80kHz; replace the reference channel Ref_ch in turn and calculate the average value of the cross-correlation coefficient R between it and other channels Ref_ch_avg , and judge:

[0098] If the current channel R Ref_ch_avg >Excellent threshold, indicating that the channel is basically normal and can be used as a reference channel. The device will then enter the fast self-test mode, which takes about 2 minutes to obtain statistical results and charts.

[0099] If we traverse all channels R Ref_ch_avg If all are ≤ the excellent threshold, it indicates that no qualified reference channel Ref_ch is found, and the device will enter the full-channel cross-check mode. It takes about 2 hours to obtain the statistical results and charts.

[0100] It prompts whether the equipment can be used normally, whether it needs maintenance, and ends the self-test.

[0101] The specific steps are:

[0102] 1) In a normal indoor / outdoor noise environment (background sound about 40–65dB (A-weighted)), the acoustic camera enters the self-test mode and controls the ultrasonic scanner via RS232 serial communication.

[0103] "20k–80kHz@FM step 1kHz@time step 1s" automatically sweeps the frequency, completing one sweep in 60s.

[0104] 2) Each microphone in the microphone array responds to the swept sound wave simultaneously, and the FPGA collects the waveform and packages the data; the waveform data is transmitted to the computing unit of the acoustic camera (usually a combination of two or more of the CPU, ARM, GPU, and FPGA).

[0105] 3) Calculate the Welch power spectral density (PSD) of all channels according to the Welch method formula (1). In order to avoid the interference of ambient audible sound, high-frequency noise, and microphone DC component, it is necessary to intercept the power spectral density (PSD) between 20k–80kHz. 20k–80k ).

[0106] 4) Fast self-test mode: Initialize the reference channel Ref_ch = 1, and calculate the power spectrum density PSD of the Ref_ch channel according to formula (2) 20k–80k Power spectral density PSD of all channels 20k–80k The result is a one-dimensional matrix with the matrix elements rounded to two decimal places. The correlation coefficient of the reference channel Ref_ch to itself (i.e., the value 1) is assigned to "empty" (i.e., it is indirectly eliminated and does not participate in subsequent calculations and plotting). The final result is a one-dimensional matrix (abbreviated as R Ref_ch ), as shown in the statistical results and charts in Table 2.

[0107] 5) Calculate the one-dimensional matrix R Ref_ch The average value of all elements of R Ref_ch_avg , make the following judgment and processing:

[0108] If R Ref_ch_avg ≤ the excellent threshold (usually about 0.8), indicating that this channel is not suitable as a reference channel and needs to be replaced with the next channel, Ref_ch+1, and repeat step 4).

[0109] If we traverse all channels R Ref_ch_avg If all are ≤ the excellent threshold (usually about 0.8), a prompt message such as "No qualified reference channel Ref_ch found. It may be that there are many damaged microphones or the array is faulty. The device will perform a full-channel cross-check, which will take about 2 hours. Please wait patiently! Or shut down the device and contact the manufacturer for after-sales service!" will be displayed on the screen (of the acoustic camera), and the process will jump to step 8).

[0110] If R Ref_ch_avg >Excellent threshold (usually about 0.8), indicating that the channel is basically normal and can be used as a reference channel; usually a qualified reference channel is quickly found, and then proceed to step 6).

[0111] 6) According to the health classification interval and threshold Table 2, the correlation coefficient one-dimensional matrix R is calculated respectively Ref_ch In the , the number, proportion, channel number, and position coordinates of the elements in each interval.

[0112] 7) Processing and presentation of statistical results:

[0113] (1) In the built-in configuration file, the serial numbers of degraded and failed channels are recorded and automatically eliminated from subsequent calculations, thereby avoiding affecting the sensitivity and positioning accuracy of the entire instrument.

[0114] (2) Calculate the total proportion of failure intervals and decay intervals. If it is greater than 15% (the maintenance thresholds for different formations may vary slightly), remind the user to perform maintenance in time to prevent the fault from expanding.

[0115] (3) Figure 12 – Figure 15 As shown, it is displayed on the (acoustic camera) screen in a graphic and textual manner (green indicates qualified, orange indicates decay, magenta indicates degradation, and red indicates failure). It is also stored in the acoustic camera and can be exported to a PC using a data cable or USB flash drive.

[0116] 8) Full channel cross self-test mode: Calculate the power spectrum density PSD of all channels according to formula (2) 20k–80k Cross-over all channel power spectral density PSD 20k–80k The result is a two-dimensional matrix with the matrix elements rounded to two decimal places. The elements on the main diagonal, i.e. the correlation coefficient of each channel to itself (i.e. the value 1), are assigned the value "empty" (i.e., indirectly eliminated and not involved in subsequent calculations and drawing). The final result is a two-dimensional matrix (abbreviated as R All_ch ); then calculate the average value of each row, round it to two decimal places, and form a new one-dimensional matrix (abbreviated as R All_ch_avg );R All_ch 、R All_ch_avgt As shown in Table 3.

[0117] 9) According to the classification interval and threshold of health degree in Table 1, the correlation coefficient one-dimensional matrix R is calculated respectively. All_ch_avg In the , the number, proportion, channel number, and position coordinates of the elements in each interval.

[0118] 10) Processing and presentation of statistical results:

[0119] (1) In the built-in configuration file, the serial numbers of degraded and failed channels are recorded and automatically eliminated from subsequent calculations, thereby avoiding affecting the sensitivity and positioning accuracy of the entire instrument.

[0120] (2) Calculate the total proportion of failure intervals and decay intervals. If it is greater than 15% (the maintenance thresholds for different formations may vary slightly), remind the user to perform maintenance in time to prevent the fault from expanding.

[0121] (3) Figure 16 – Figure 18 As shown, the results are displayed on the (acoustic camera) screen in a graphic and textual manner (green indicates qualified, orange indicates decay, magenta indicates degradation, and red indicates failure). They are also stored in the acoustic camera and can be exported to a PC using a data cable or USB flash drive.

[0122] In summary:

[0123] (1) Steps 4)–7) are the “Quick Self-Test Mode”. By automatically searching for qualified reference channels and calculating the one-dimensional correlation coefficient matrix between the reference channels and all channels, statistical results and charts can be obtained in about 2 minutes, completing the preliminary screening of abnormal channels to meet daily operation and maintenance needs.

[0124] (2) Steps 8)–10) are the “full-channel cross-check mode”. By first constructing a two-dimensional correlation coefficient matrix and then performing average value analysis on each row to improve the detection robustness, the results are more accurate, but it takes slightly longer, about 2 hours to obtain statistical results and charts. It is suitable for equipment production and maintenance self-checking. The sweep frequency range can be further expanded from 20k–80kHz to 1k–100kHz.

[0125] (3) Due to the bias in the reference channel Ref_ch selected in the "Quick Self-Test Mode," the correlation coefficient values ​​and result distributions of the two modes may differ slightly. However, the results of the "Qualified Interval + Degradation Interval" and "Degradation Interval + Failure Interval" are essentially identical. The self-test focuses on the distribution of the "Degradation Interval + Failure Interval." It is entirely feasible to use the "Quick Self-Test Mode" for routine operation and maintenance.

[0126] Software algorithms and formulas:

[0127] Array systems typically use dozens to hundreds of MEMS microphones. Due to the different spatial locations of the microphones, the acoustic signals they collect inevitably exhibit time delay and phase differences. A properly performing microphone should collect signals with consistent frequency components and exhibit similar morphological characteristics in its power spectral density (PSD). Therefore, the correlation coefficient of the power spectral density can be used to assess the health of the microphones, enabling the detection and screening of abnormal channels.

[0128] The Welch method has become one of the most commonly used power spectrum estimation methods in engineering practice due to its good comprehensive performance. The basic steps are: first, segment the random sequence so that each segment has some overlap (usually 50%), then smooth each segment of data with a suitable window function, and finally average the spectra of each segment to obtain the power spectrum.

[0129] Divide the signal X(j) of length N into K segments, each segment is of length L, and the overlap between adjacent segments is D, usually D = L / 2, such as Figure 1 shown.

[0130] N=L+(K-1)(LD), and the solution is

[0131] Apply the Hanning window function W(j) to each signal segment and reduce spectrum leakage

[0132]

[0133] Calculate the discrete Fourier transform (DFT) of each windowed segment to convert the time domain signal to the frequency domain

[0134] in

[0135] Get the modified periodogram of segment K

[0136] The frequency is discretized into f s is the sampling rate, n=0,1,…,L / 2

[0137] The normalization factor U is used to compensate for the energy loss caused by windowing

[0138] All K periodograms are averaged to obtain the final Welch power spectral density estimate

[0139] The combined formula is

[0140] Correlation coefficient

[0141] The correlation coefficient is used to measure the strength and direction of the linear relationship between two variables (X and Y). Its value range is between [-1, 1]. 1: Perfect positive correlation (as X increases, Y also increases linearly).

[0142] -1: Perfect negative correlation (as X increases, Y decreases linearly).

[0143] 0: No linear correlation (but there may be nonlinear relationships).

[0144] According to the spectrum estimation method, the autopower spectrum of the two microphones is obtained by formula (1): Calculate the correlation coefficient of two channels

[0145]

[0146] Where cov represents and The covariance between They are The standard deviation of .

[0147] The classification intervals and thresholds of healthiness are shown in Table 1.

[0148]

[0149] Table 1

[0150] The following is a further explanation and verification with specific examples.

[0151] Example 1: On a sunny roadside, with a constant breeze, occasional pedestrians talking and passing by, and occasional cars honking and passing by, the ambient noise level was 53.9–65.4 dB (A-weighted values ​​measured by a SanLiang SM550 sound level meter). An acoustic camera with 120 MEMS microphone channels was used to perform a health self-assessment.

[0152] Steps 4)–7) are in "Quick Self-Test Mode" and take about 2 minutes to obtain the statistical results and charts for 120 microphone channels. Table 2 shows the one-dimensional matrix of the correlation coefficients of the reference channel Ref_ch to the power spectral density of all channels.

[0153]

[0154] Table 2

[0155] (Note: ··· indicates omission; reference channel Ref_ch = 1, so the intersection with ch1 is assigned a value of "empty")

[0156] As can be seen from Table 2, ch16 corresponds to 0, indicating that the microphone is out of service; ch44 corresponds to 0.54, and ch74 corresponds to 0.67, which are in the range of 0.5–0.7, indicating that the microphone is degraded.

[0157] Figure 14 – Figure 18 In the chart, green indicates passing, orange indicates decay, magenta indicates degradation, and red indicates failure.

[0158] from Figure 14 、 Figure 15 As can be seen from the figure, the quick self-test results of the 120 microphone channels are as follows: 117 are qualified, accounting for 97.5%; microphones ch44 and ch74 are degraded, accounting for 1.7%; 0 channels are degraded, accounting for 0%; microphone ch24 is invalid, accounting for 0.8%; the total degradation failure ratio = 0% + 0.8% = 0.8% < 15% (repair threshold), and the equipment can be used normally / no maintenance is required.

[0159] Steps 8)–10) are in the "full channel cross-check mode," and it takes about 2.5 hours for 120 microphone channels to obtain statistical results and charts. Table 3 shows the two-dimensional matrix of correlation coefficients of the cross-correlation of all channel power spectral densities to all channel power spectral densities.

[0160]

[0161] Table 3

[0162] (Note: ··· indicates omission; the intersection of channel rows and columns (correlation coefficient of each channel with itself) is assigned a value of "blank")

[0163] From the R in Table 3 All_ch_avg It can be seen from the column that ch16 corresponds to 0, indicating that the microphone fails; ch44 corresponds to 0.53, and ch74 corresponds to 0.62, which are in the range of 0.5–0.7, indicating that the microphone is degraded.

[0164] from Figure 16 、 Figure 17 、 Figure 18 As can be seen from the figure, the results of the full-channel cross-interactive self-test of 120 microphones are as follows: 117 are qualified, accounting for 97.5%; microphones ch44 and ch74 are degraded, accounting for 1.7%; 0 channels are degraded, accounting for 0%; microphone ch16 is invalid, accounting for 0.8%; the total degradation failure ratio = 0% + 0.8% = 0.8% < 15% (repair threshold), and the equipment can be used normally / no maintenance is required.

[0165] In summary, the results of the "quick self-test mode" and the "full-channel cross self-test mode" are basically the same.

[0166] Example 2: In an office environment with occasional movement and conversation in the room, and voices from a conference room next door, the ambient noise level is 37.3–59.3 dB (A-weighted value measured by a SanLiang SM550 sound level meter). An acoustic camera with 60 MEMS microphone channels is used to perform a health status self-assessment.

[0167] Steps 4)–7) are in "Quick Self-Test Mode" and take about 2 minutes to obtain the statistical results and charts for 60 microphone channels. Table 4 shows the one-dimensional matrix of the correlation coefficients of the reference channel Ref_ch to the power spectral density of all channels.

[0168]

[0169] Table 4

[0170] (Note: ··· indicates omission; reference channel Ref_ch = 1, so the intersection with ch1 is assigned a value of "empty")

[0171] From Table 4, we can see that ch40 corresponds to 0.6, which is in the range of 0.5–0.7, indicating microphone attenuation.

[0172] Figure 22 – Figure 26 In the chart, green indicates passing, orange indicates decay, magenta indicates degradation, and red indicates failure.

[0173] from Figure 22 、 Figure 23As can be seen from the figure, the quick self-test results of 60 microphone channels are as follows: 59 are qualified, accounting for 98.3%; ch40 microphones are degraded, accounting for 1.7%; 0 channels are degraded, accounting for 0%; 0 channels are failed, accounting for 0%; the total degradation failure ratio = 0% + 0% = 0% < 15% (maintenance threshold), and the equipment can be used normally / no maintenance is required.

[0174] Steps 8)–10) are in the "full channel cross-check mode." It takes about two hours for the 120 microphone channels to obtain statistical results and charts. Table 5 shows the two-dimensional matrix of correlation coefficients of the cross-correlation of the power spectral density of all channels to the power spectral density of all channels.

[0175]

[0176] Table 5

[0177] (Note: ··· indicates omission; the intersection of channel rows and columns (correlation coefficient of each channel with itself) is assigned a value of "blank")

[0178] From the R in Table 5 All_ch_avg It can be seen from the column that ch40 corresponds to 0.55, which is in the range of 0.5–0.7, indicating microphone attenuation.

[0179] from Figure 24 、 Figure 25 、 Figure 26 As can be seen from the figure, the results of the full-channel cross-interactive self-test of 60 microphones are as follows: 59 are qualified, accounting for 98.3%; ch40 microphones are degraded, accounting for 1.7%; 0 channels are degraded, accounting for 0%; 0 channels fail, accounting for 0%; the total degradation failure ratio = 0% + 0% = 0% < 15% (maintenance threshold), and the equipment can be used normally / no maintenance is required.

[0180] In summary, the results of the "quick self-test mode" and the "full-channel cross self-test mode" are basically the same.

Claims

1. A method for in-situ quantitative evaluation of the health status of a MEMS microphone array in a noisy environment, characterized in that: The steps include: The acoustic camera enters the self-test mode and controls the ultrasonic frequency scanner through communication to perform automatic frequency scanning. Each microphone in the microphone array responds to the swept frequency sound wave simultaneously. The FPGA collects the waveform and packages the data before transmitting it to the computing unit of the acoustic camera. Calculate the Welch power spectral density of all channels and intercept the power spectral density between 20k-80kHz; Replace the reference channel Ref_ch in turn, calculate the average value of the correlation coefficient RRef_ch_avg between it and other channels, and determine: If the current channel R Ref_ch_avg >Excellent threshold, indicating that this channel is basically normal and can be used as a reference channel, then the device will enter the fast self-test mode, and it takes a total of 2 minutes to obtain statistical results and charts; If we traverse all channels R Ref_ch_avg If all are ≤ the excellent threshold, it indicates that no qualified reference channel Ref_ch is found, and the device will enter the full-channel cross-check mode. It takes a total of 2 hours to obtain the statistical results and charts. It prompts whether the equipment can be used normally, whether it needs maintenance, and ends the self-test.

2. The in-situ quantitative assessment method for the health status of a MEMS microphone array in a noisy environment according to claim 1, characterized in that: The quick self-test mode includes the following steps: S1) Initialize the reference channel Ref_ch = 1, and calculate the power spectrum density PSD of the Ref_ch channel according to formula (2) 20k–80k Power spectral density PSD of all channels 20k–80k The result is a one-dimensional matrix with the matrix elements rounded to two decimal places. Then the correlation coefficient of the reference channel Ref_ch to itself is assigned to "null". The final result is a one-dimensional matrix. S2) Calculate the one-dimensional matrix R Ref_ch The average value of all elements of R Ref_ch_avg , make the following judgment: If R Ref_ch_avg ≤ the excellent threshold, indicating that the current channel is not suitable as a reference channel and needs to be replaced with the next channel, Ref_ch+1, and then repeat step S1); If we traverse all channels R Ref_ch_avg If all the values ​​are less than or equal to the excellent threshold, a prompt message will be displayed on the screen of the acoustic camera, and the system will switch to the full-channel cross-check mode. If R Ref_ch_avg >Excellent threshold, indicating that the channel is basically normal and can be used as a reference channel; usually a qualified reference channel is quickly found, and the process goes to step S3; S3) According to the classification interval and threshold of health degree, the correlation coefficient one-dimensional matrix R is calculated respectively Ref_ch The number, proportion, channel number, and position coordinates of the elements in each interval; S4) Processing and presentation of statistical results.

3. The in-situ quantitative assessment method for the health status of a MEMS microphone array in a noisy environment according to claim 1, characterized in that: The full channel cross-check mode includes the following steps: Step 1) Calculate the power spectral density PSD of all channels according to formula (2) 20k–80k Cross-over all channel power spectral density PSD 20k–80k The result is a two-dimensional matrix, with the matrix elements rounded to two decimal places; the elements on the main diagonal are assigned to "empty"; the final result is a two-dimensional matrix; the average value of each row is calculated, rounded to two decimal places, and a new one-dimensional matrix is ​​formed; step 2) the correlation coefficient one-dimensional matrix R is calculated according to the classification interval and threshold of healthiness. All_ch_avg The number, proportion, channel number, and position coordinates of the elements in each interval; Step 3) How to process and present the statistical results.

4. The in-situ quantitative assessment method for the health status of a MEMS microphone array in a noisy environment according to claim 1, characterized in that: Divide the signal X(j) of length N into K segments, each of length L, with overlap D between adjacent segments, usually D = L / 2; N=L+(K-1)(LD), and the solution is Multiply each signal segment by the Hanning window function W(j) and reduce spectrum leakage Calculate the discrete Fourier transform of each windowed segment to convert the time domain signal to the frequency domain in Get the modified periodogram of segment K The frequency is discretized into f s is the sampling rate, n=0,1,…,L / 2 The normalization factor U is used to compensate for the energy loss caused by windowing All K periodograms are averaged to obtain the final Welch power spectral density estimate The combined formula is 5. The in-situ quantitative evaluation method for the health status of a MEMS microphone array in a noisy environment according to claim 4, characterized in that: According to the spectrum estimation method, the autopower spectrum of the two microphones is obtained by formula (1): Calculate the correlation coefficient of two channels Among them, cov represents and The covariance between They are The standard deviation of .

6. A device for the in-situ quantitative evaluation method of the health status of a MEMS microphone array in a noisy environment according to any one of claims 1 to 5, characterized in that: The device includes an embedded ultrasonic scanner including: The power conversion module is a DC / DC power conversion module; the power supply and communication module is connected to the power conversion module; The communication control and waveform generation MCU is connected to the power conversion module, and the power supply and communication module is connected to the communication control and waveform generation MCU; The integrated amplifier chip is connected to the communication control and waveform generation MCU, and the integrated amplifier chip is connected to the power conversion module; Ultrasonic speaker, connected to the integrated amplifier chip.

Images