An acoustic online non-destructive testing method and system for a flight of a flight of a flight of a flight of a flight of a flight of a flight of a

Abstract

The application relates to the technical field of coal mine fully-mechanized coal mining face transportation equipment monitoring, and discloses a kind of acoustic on-line nondestructive testing method and system for scraper conveyor scraper, the method converts the PDM signal collected into PCM data using cascaded integral comb filter;The power spectral density is calculated by Welch method, the characteristic vector is constructed by sub-band energy integration, L2 norm normalization and logarithmic transformation;Fault determination is carried out using cosine similarity, and sound source positioning is realized by combining generalized cross-correlation and hyperboloid intersection method.The system updates the voiceprint library using the weighted moving average algorithm based on artificial review data, and issues parameters in real time through the double-buffer hot loading mechanism.The hardware adopts zero-light-window sealed structure and honeycomb shock-absorbing seat, effectively isolating environmental noise and structural vibration.The application solves the problems of difficulty in extracting scraper fault features and inaccurate positioning in a strong noise environment, and improves the detection accuracy and environmental adaptability.

Description

Technical Field

[0001] This invention relates to the field of monitoring technology for transportation equipment in fully mechanized coal mining faces, specifically to an online non-destructive testing method and system for the acoustics of scraper conveyors. Background Technology

[0002] As coal mining progresses to deeper levels, scraper conveyors, as key transportation equipment in fully mechanized mining faces, directly determine the mine's production capacity and safety through their operational reliability. Within the scraper chain assembly, the scrapers not only bear the functions of pushing coal and carrying loads, but also endure multiple combined conditions of impact, abrasive wear, and corrosive media such as water, hydrogen sulfide, and coal slime over extended periods, making them one of the most vulnerable components with a high failure rate. Statistics show that unplanned downtime due to scraper breakage or excessive wear accounts for a significant portion of the total downtime for scraper conveyors. Furthermore, chain breakage in narrow, high-gas roadways can easily trigger coal flow blockages, motor burnout, and even secondary gas accidents.

[0003] Currently, monitoring the quality of scraper blades in underground coal mines mainly relies on manual visual inspection by maintenance technicians. Existing technologies have also seen some monitoring solutions based on contact sensing or single optical measurement methods. One such solution utilizes a cylinder fixed to the frame to drive a pressure sensor at the lower end of the telescopic arm to contact the chain, combined with single-point laser measurement of the scraper blade position.

[0004] However, in practical applications, the aforementioned detection methods are susceptible to damage due to the severe vibrations encountered during the operation of scraper conveyors. The mechanical structures and pressure sensors used in contact monitoring are prone to damage under prolonged high-frequency vibration, leading to insufficient overall stability of the monitoring equipment. Furthermore, the underground working environment is typically characterized by high concentrations of coal dust, which can easily obstruct optical measurement paths such as lasers, causing signal attenuation or drift and increasing measurement errors. This makes it difficult to accurately identify the fault location of the scraper conveyor under complex operating conditions. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an online non-destructive acoustic testing method and system for scraper conveyors, which solves the problems of poor stability, susceptibility to signal interference, and inaccurate fault identification and location of existing contact and optical monitoring methods and equipment in the context of strong vibration and high dust in underground coal mines.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method and system for online non-destructive acoustic testing of scraper conveyors.

[0007] The first aspect of this invention provides an online non-destructive testing method for the acoustic properties of scraper conveyors.

[0008] This method involves synchronously acquiring acoustic signals from the operating environment of the scraper conveyor using a sound acquisition layer, and converting the output pulse density modulation signal into pulse code modulation data. During the conversion process, multiple pulse density modulation signals are received, and a cascaded integrator-comb filter algorithm is used for downsampling and high-frequency quantization noise removal, converting the 1-bit data stream into 24-bit linear audio data. The effective sampling rate of the downsampling is calculated by dividing the driving clock frequency by the downsampling factor. The data is uploaded after adding the array number and timestamp.

[0009] The signal processing layer performs bandpass filtering, overlapping framing, and time-domain windowing on the received data. The bandpass filtering uses a finite-length unit impulse response filter with a frequency range of 1 kHz to 8 kHz. Time-domain windowing is performed on each frame of the signal using the Hamming window function formula. Subsequently, a frequency domain transformation is performed on the windowed signal, and the power spectral density is calculated by accumulating and averaging the periodograms of multiple consecutive frames using the Welch averaging method.

[0010] In the feature extraction stage, the energy integral of the power spectral density within the divided key frequency sub-bands is calculated using a numerical integration algorithm. The energy integral results are then normalized using the L2 norm to eliminate signal strength differences, and the fault feature vector is generated using the natural logarithmic transform method.

[0011] During the fault determination phase, the cosine similarity formula is used to calculate the cosine similarity between the fault feature vector and the centroid vector in the stored fault soundprint database. When the similarity is greater than the baseline determination threshold, a fault is determined to exist. This step includes confidence leveling logic: if the cosine similarity is greater than or equal to the minimum determination threshold but less than the high confidence threshold, a verification status code is generated, and the audible and visual alarm emits a yellow warning light; if the cosine similarity is greater than or equal to the high confidence threshold, a high confidence status code is generated, and the audible and visual alarm emits a red flashing light and intermittent buzzing sound.

[0012] The method also includes a sound source localization step: selecting microphone signals from different array nodes to construct cross-correlation pairs, calculating the cross-correlation function using a generalized cross-correlation algorithm, searching for the peak position of the cross-correlation function, and determining the time delay of the sound wave arriving at different arrays. Based on the distance difference principle, a system of equations for a hyperboloid of revolution with the microphone position as the focus is constructed, and the three-dimensional spatial coordinates of the sound source are calculated by solving the system of equations for a hyperboloid of revolution using the least squares method and iterative method.

[0013] The system updates the model through a closed-loop feedback mechanism: it receives manually reviewed data returned by the feedback decision layer, and the mobile terminal displays the rack number of the fault point and receives near-field audio recorded and photos taken by maintenance personnel. If the review result confirms the fault, the original array sound data is marked as a positive sample. When updating the fault acoustic signature database, the positive sample data set is extracted, and the average feature vector of the current batch of samples is calculated using an arithmetic mean algorithm. Using a weighted moving average formula, based on the configured update factor, the weighted sum of the old centroid vector and the current batch's average feature vector is calculated to obtain the updated centroid vector.

[0014] The updated parameters are distributed through a double-buffered hot-loading mechanism: the signal processing layer receives the configuration file and uses the MD5 digest algorithm to calculate the MD5 value for integrity verification; after the verification is successful, the parameters are written into the loading area in memory; during the interval between the processing of the current frame data, the memory pointer is switched through atomic operation methods to mark the loading area as active, so that the updated parameters can be called in the next frame data.

[0015] A second aspect of the present invention provides an online non-destructive testing system for the acoustic properties of a scraper conveyor, comprising a sound acquisition layer, a signal processing layer, and a feedback decision layer.

[0016] The sound acquisition layer is arranged longitudinally along the scraper conveyor body, including first, second, and third MEMS microphone arrays respectively located at the head, tail, and middle of the body, as well as a sound acquisition module and a sound transmission and numbering storage module. The signal processing layer includes a signal processing module connected to the sound acquisition layer via a bus. The feedback decision layer includes a monitoring host, mobile terminal, and remote server connected to the signal processing module via an underground industrial ring network.

[0017] The MEMS microphone array adopts an array-type sensor node structure, including a cast aluminum explosion-proof housing. The front end of the housing is covered with a thin-walled metal cover, forming a zero-light-window sealed structure. Inside the housing are multiple microelectromechanical system microphone units arranged in a circular pattern. The housing is fixed by a composite vibration damping mounting mechanism, which includes a silicone vibration damping seat with a honeycomb-shaped vibration damping hole array, and a metal limiting sleeve is embedded in the vibration damping seat.

[0018] This invention provides a method and system for online non-destructive acoustic testing of scraper conveyors. It offers the following advantages: 1. This invention utilizes a zero-light-window sealed structure formed by a cast aluminum explosion-proof shell and a front-end thin-walled metal cover. This structure physically blocks the intrusion of dust and water vapor from the well, while simultaneously transmitting sound waves using the vibration and sound transmission characteristics of the thin metal plate. Combined with a honeycomb-shaped silicone vibration damping seat with an embedded metal limiting sleeve, the honeycomb array reduces radial stiffness, effectively isolating low-frequency structural vibrations transmitted from the scraper conveyor body. This ensures that the internally distributed circumferentially distributed MEMS microphone units primarily pick up acoustic emission signals transmitted through the air, suppressing mechanical vibration interference from the hardware source.

[0019] 2. This invention employs the Welch averaging method to obtain a stable power spectral density and calculates the energy integral of key frequency sub-bands. It utilizes the L2 norm normalization formula to eliminate differences in absolute signal intensity caused by source distance or scraper load fluctuations, and combines this with natural logarithmic transformation to map features to a linear space, constructing a fault feature vector sensitive only to frequency distribution patterns. Cosine similarity is used instead of traditional Euclidean distance for comparison, focusing on differences in vector direction, thereby accurately identifying fault characteristics of the scraper chain under complex operating conditions.

[0020] 3. This invention utilizes manually reviewed data returned from the feedback decision layer as positive samples. An update factor is calculated using a weighted moving average algorithm to dynamically correct the centroid vector of the voiceprint database, enabling the detection system to adapt to voiceprint drift caused by wear of the scraper and chute. Combined with a double-buffered hot-loading mechanism, new parameters are loaded in memory via atomic operations by switching pointers, achieving real-time updates to the detection model without interrupting signal acquisition and processing tasks, thus ensuring the system's continuous monitoring capability. Attached Figure Description

[0021] Figure 1 This is a block diagram of the overall system architecture of the present invention; Figure 2 This is a schematic diagram of the installation layout of the system of the present invention on a scraper conveyor; Figure 3 This is the main flowchart of the online non-destructive testing method for scraper quality of the present invention; Figure 4 This is a flowchart of the signal acquisition and preprocessing process of the present invention; Figure 5 This is an acoustic graph of the sound emitted by the scraper movement during normal operation of the scraper conveyor of the present invention. Figure 6 This is a frequency diagram of the sound emitted by the scraper movement during normal operation of the scraper conveyor of the present invention; Figure 7 This is an acoustic waveform diagram of the sound emitted by the scraper movement when the scraper conveyor of the present invention malfunctions. Figure 8 This is a frequency diagram of the sound emitted by the scraper movement when the scraper conveyor of the present invention malfunctions. Figure 9 This is a flowchart of the fault alarm and monitoring data reporting process of the present invention; Figure 10 This is a flowchart illustrating the mobile terminal interaction and manual review process of the present invention. Figure 11 This is a flowchart of the incremental training and OTA online upgrade process of the model in this invention.

[0022] Among them, 10 is a cast aluminum explosion-proof housing; 101 is a first MEMS microphone array; 102 is a second MEMS microphone array; 103 is a third MEMS microphone array; 104 is a sound acquisition module; 105 is a sound transmission and numbering storage module; 11 is a silicone shock-absorbing base; 12 is an ear base; 13 is a thin-walled metal cover; 200 is a signal processing module; 301 is a monitoring host; 302 is a mobile terminal; and 303 is a remote server. Detailed Implementation

[0023] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Please see the appendix Figure 1 and attached Figure 2 This invention provides a method and system for online non-destructive acoustic testing of scraper conveyors, comprising a sound acquisition layer, a signal processing layer, and a feedback decision layer.

[0025] The sound acquisition layer is arranged longitudinally along the body of the scraper conveyor at the tunneling face, and is used to pick up acoustic emission signals during the operation of the scraper chain. The sound acquisition layer includes a first MEMS microphone array 101 located in the head region, a second MEMS microphone array 102 located in the tail region, and a third MEMS microphone array 103 located in the middle region of the body. The first MEMS microphone array 101, the second MEMS microphone array 102, and the third MEMS microphone array 103 are all connected to a sound acquisition module 104. The sound acquisition module 104 is connected to a sound transmission and numbering storage module 105.

[0026] The signal processing layer includes a signal processing module 200, which is connected to the sound transmission and number storage module 105 via an RS-485 differential bus. The signal processing module 200 integrates a digital signal processing unit for performing time-frequency analysis, feature extraction, and fault diagnosis on the received acoustic data.

[0027] The feedback decision-making layer includes a monitoring host 301, a mobile terminal 302, and a remote server 303. The output of the signal processing module 200 is connected to the monitoring host 301, mobile terminal 302, and remote server 303 via the underground industrial ring network. The monitoring host 301 is located in the underground roadway control room and is used to display real-time monitoring status and alarm information. The mobile terminal 302 is configured as a handheld explosion-proof device for maintenance personnel. The remote server 303 is located in the surface data center and is used to store historical data and update algorithm models.

[0028] Please see the appendix Figure 3 The online non-destructive acoustic testing method for scraper conveyors according to this embodiment includes the following steps: S1. Sound Signal Acquisition and Numbering. The first MEMS microphone array 101, the second MEMS microphone array 102, and the third MEMS microphone array 103 synchronously acquire sound wave signals from the operating environment of the scraper conveyor and output pulse density modulated signals. The sound acquisition module 104 receives multiple pulse density modulated signals, performs clock alignment and downsampling processing on them, and converts them into pulse code modulation data. The sound transmission and numbering storage module 105 adds an array number and a millisecond-level timestamp to each frame of pulse code modulation data and uploads it to the signal processing module 200 via the bus.

[0029] S2. Audio signal time-frequency filtering. After receiving the data frame, the signal processing module 200 uses a built-in finite-length unit impulse response bandpass filter to filter the signal. The passband range of the filter is set to 1 kHz to 8 kHz to remove low-frequency environmental noise and high-frequency mechanical aliasing noise. Subsequently, the signal processing module 200 performs overlapping frame segmentation on the filtered signal and applies a Hamming window to each frame.

[0030] S3. Calculation of audio signal spectrum. The signal processing module 200 performs a fast Fourier transform on each frame of the windowed time-domain signal, calculates the amplitude spectrum of the signal, and converts the time-domain data into frequency-domain data.

[0031] S4. Solving for signal spectral power density and feature extraction. The signal processing module 200 calculates the power spectrum of a single frame based on the amplitude spectrum, and uses the Welch averaging method to accumulate and average the power spectra of multiple consecutive frames to obtain a smoothed power spectral density. Subsequently, the signal processing module 200 calculates the energy integral of the power spectral density within several preset key frequency sub-bands, and performs L2 norm normalization and logarithmic transformation on the energy integral result to generate a fault feature vector.

[0032] S5. Compare with a preset voiceprint library. The signal processing module 200 reads a preset fault voiceprint library from its internal memory. This library contains centroid vectors for various fault types. The signal processing module 200 calculates the cosine similarity between the real-time generated fault feature vector and each centroid vector. When the calculated cosine similarity value is greater than a preset judgment threshold, it is determined that the current scraper chain has a fault of the corresponding type, and a fault type code is generated.

[0033] S6. Determining Fault Location Using Multi-Channel Signals. Upon determining the presence of a fault, the signal processing module 200 utilizes multi-channel signals from the first MEMS microphone array 101, the second MEMS microphone array 102, and the third MEMS microphone array 103 to perform generalized cross-correlation calculations to estimate the time delay of the sound source reaching each array. The signal processing module 200 then combines the geometric coordinates of each array and uses the hyperboloid intersection method to calculate the three-dimensional spatial coordinates of the sound source.

[0034] S7. Output quality assessment results and closed-loop feedback. The signal processing module 200 sends a data packet containing the fault type code, three-dimensional spatial coordinates, and timestamp to the feedback decision layer. After receiving the data packet, the monitoring host 301 marks the fault location on the graphical interface and triggers an alarm. The remote server 303 records the detection result. When maintenance personnel receive instructions via the mobile terminal 302 and complete on-site verification, they send the verification results back to the remote server 303. The remote server 303 uses the verification data to update the fault acoustic signature database and sends the updated parameters to the signal processing module 200 via online upgrade.

[0035] In this embodiment, the sound acquisition layer consists of several independent array-type sensor nodes with the same structure. These nodes correspond to the first MEMS microphone array 101, the second MEMS microphone array 102 and the third MEMS microphone array 103 of Embodiment 1, respectively.

[0036] The array-type sensor node includes a cast aluminum explosion-proof housing 10, which is made of high-strength cast aluminum alloy, and its mechanical strength meets the pressure resistance and impact resistance requirements of explosion-proof electrical equipment enclosures in underground coal mines. The cast aluminum explosion-proof housing 10 has an internal cavity for installing the acoustic-to-electrical conversion components and signal conditioning circuitry. The sidewalls of the cast aluminum explosion-proof housing 10 are provided with lug structures for fixed installation, which are adapted to the support beam of the scraper conveyor.

[0037] Inside the cavity, an acoustic acquisition circuit board assembly is disposed. This assembly employs a multi-layer printed circuit board stack structure, including a microphone array board at the bottom and a signal gathering board at the top. The microphone array board has a circular structure, with multiple MEMS microphone units evenly distributed along its circumference in its edge region. In this embodiment, there are six MEMS microphone units, denoted as microphone 101a to microphone 101f. These six microphones are located on a circle with the same radius RR, centered on the geometric center of the microphone array board, and the central angle between any two adjacent microphones is equal, i.e., the central angle is 60°. This uniformly distributed circumferential geometric topology provides an isotropic reference plane for subsequent sound source localization based on the time difference of arrival algorithm.

[0038] The MEMS microphone units utilize high signal-to-noise ratio digital output sensors with an acoustic overload point of no less than 120 dB sound pressure level to adapt to the high sound pressure conditions in the scraper conveyor operating environment. Each MEMS microphone unit is surface-mounted to the top surface of the microphone array board, with the sound inlet of each microphone facing the front end of the cast aluminum explosion-proof housing 10. The microphone array board integrates power supply filtering circuitry and impedance matching circuitry to provide a stable operating voltage for the MEMS microphone units and ensure the integrity of the digital pulse signals. The specific implementation of the surface-mount soldering process and the basic filtering circuitry is well-known in the field and will not be described in detail here.

[0039] The signal aggregation board is electrically connected to the microphone array board via board-to-board connectors or flexible ribbon cables. The signal aggregation board integrates a field-programmable gate array (FPGA) or a high-performance microcontroller, serving as the hardware carrier for the aforementioned sound acquisition module 104. The signal aggregation board is responsible for providing synchronization clock signals to the six MEMS microphone units, ensuring that the acoustic signals of all channels are strictly aligned on the time axis. The signal aggregation board also includes an RS-485 communication interface circuit and a power management circuit. The communication interface circuit is led out through shielded twisted-pair cables to a wiring cavity outside the cast aluminum explosion-proof housing 10, enabling data interaction with the signal processing module 200.

[0040] The cast aluminum explosion-proof housing 10 does not have through holes directly connecting to the external environment at the location corresponding to each MEMS microphone unit. Instead, it receives sound through a special acoustic coupling structure, the specific structure of which will be detailed in subsequent embodiments. This hardware construction of the array-type sensor node integrates the acoustic-to-electric conversion unit, the signal digitization unit, and the explosion-proof protection structure into a single module, achieving high-fidelity, multi-channel synchronous acquisition of the operating noise of the scraper conveyor.

[0041] In this embodiment, the array-type sensor node adopts a sealed encapsulation structure without light-transmitting holes or ventilation holes, i.e., a zero-light-window design, to adapt to the harsh working conditions of high dust, water spray, and oil sludge adhesion in underground coal mines.

[0042] The front face of the cast aluminum explosion-proof housing 10 has a circular acoustic wave receiving window, the diameter of which is larger than the diameter of the circumference of each MEMS microphone unit on the microphone array board. A thin-walled metal cover 13 is provided to cover the acoustic wave receiving window. The edge of the thin-walled metal cover 13 is fixed to the front face of the cast aluminum explosion-proof housing 10 by a sealing assembly and a clamping flange, thereby physically isolating the internal cavity of the cast aluminum explosion-proof housing 10 from the external environment.

[0043] The thin-walled metal cover 13 is made of a metal material with a certain elastic modulus and corrosion resistance. In this embodiment, the thin-walled metal cover 13 is made of 304 or 316L austenitic stainless steel sheet, with a thickness set between 0.2 mm and 0.5 mm. The selection of this thickness range is based on the balance between acoustic transmission performance and mechanical protection strength: if the thickness is less than 0.2 mm, the metal cover is prone to breakage or plastic deformation when subjected to external impact; if the thickness is greater than 0.5 mm, the surface density of the metal cover increases, which will lead to excessive sound wave transmission loss, especially attenuation of high-frequency signals where the scraper fault characteristics are located.

[0044] The sealing assembly includes a sealing gasket disposed between the contact surfaces of the thin-walled metal cover 13 and the cast aluminum explosion-proof housing 10. The sealing gasket is made of oil-resistant and aging-resistant nitrile rubber or fluororubber, and its cross-sectional shape is rectangular or O-shaped. The clamping flange is fastened to the cast aluminum explosion-proof housing 10 by bolts, applying an axial preload to the edge of the thin-walled metal cover 13, causing the sealing gasket to elastically deform and fill the mating gap, thereby achieving a sealing effect with a dustproof and waterproof rating of IP65 or higher. Contaminants such as coal dust, water spray, and emulsions can only adhere to the outer surface of the thin-walled metal cover 13 and cannot penetrate the housing to damage the MEMS microphone unit or block the sound path.

[0045] In terms of acoustic conduction, the thin-walled metal cover 13 utilizes the principle of thin-plate vibration sound transmission to achieve sound wave transmission coupling. When sound waves in the external air are incident on the surface of the thin-walled metal cover 13, the sound pressure excites the metal plate to produce forced vibration. The vibration of the metal plate then radiates sound waves into the air medium inside the cast aluminum explosion-proof housing 10, forming an internal sound field. Since the MEMS microphone unit is located in the near-field region behind the thin-walled metal cover 13, it can directly pick up the sound pressure signal of this internal sound field. This structure ensures that the sound waves only produce an insertion loss of about 3 dB when penetrating the thin-walled metal cover 13, and this loss maintains a relatively flat frequency response characteristic within the 1 kHz to 8 kHz frequency band of interest to the system, ensuring the characteristic integrity of the original acoustic emission signal. The specific acoustic model establishment and transmission coefficient calculation for thin-plate vibration sound transmission are conventional techniques in the field of acoustic engineering and will not be elaborated here.

[0046] The cast aluminum explosion-proof housing 10 is fixed to the support beam of the frame via a composite vibration damping mounting mechanism. The composite vibration damping mounting mechanism mainly includes a silicone vibration damping seat 11 disposed between the lug 12 of the cast aluminum explosion-proof housing 10 and the mounting surface of the frame.

[0047] The silicone vibration damping seat 11 is made of industrial-grade silicone rubber with high damping properties, and its Shore hardness range is set from 40 to 60 degrees. The main structure of the silicone vibration damping seat 11 is not solidly filled, but rather has a honeycomb-shaped array of damping holes. Specifically, on the load-bearing cross-section of the silicone vibration damping seat 11, several regular hexagonal through holes extending perpendicular to the mounting surface are evenly distributed. This honeycomb structure design reduces the equivalent dynamic stiffness of the damping seat, giving it non-linear deformation characteristics under compressive loads. Compared to a solid rubber block, it provides a larger deformation buffer stroke, thereby reducing the natural frequency of the entire sensor node mounting system.

[0048] For connection and fixation, the composite vibration damping mounting mechanism adopts a rigid-flexible coupling connection method. The mounting bolts are sequentially passed through the upper metal washer, the lug 12 mounting holes of the cast aluminum explosion-proof housing 10, the silicone vibration damping seat 11, and the threaded holes on the frame for tightening. To prevent excessive tightening torque of the mounting bolts from causing excessive compression and permanent deformation or loss of elasticity in the silicone vibration damping seat 11, a metal limiting sleeve is embedded in the central mounting hole of the silicone vibration damping seat 11. The axial length of the metal limiting sleeve is slightly less than the height of the silicone vibration damping seat 11 in its free state. When the bolts are tightened, the metal limiting sleeve limits the maximum compression of the silicone vibration damping seat 11, keeping it in an optimal pre-compression working state. Simultaneously, the tightening force of the bolts is mainly borne by the end face of the metal limiting sleeve, preventing vibration energy from being directly transmitted to the cast aluminum explosion-proof housing 10 through the bolt shank in a rigid contact manner.

[0049] From a dynamic perspective, the aforementioned composite vibration damping installation mechanism constructs a mechanical low-pass filter. The structural vibration energy of the scraper conveyor body is mainly concentrated in the low-frequency range, typically below 500 Hz, while the acoustic emission signal characteristics caused by faults such as scraper chain cracks and broken teeth, which are the focus of this invention, are mainly distributed in the high-frequency range of 1 kHz to 8 kHz. By selecting silicone rubber materials with specific hardness and adjusting the porosity with a honeycomb structure, the natural frequency of this vibration damping system is designed to be below 100 Hz. According to vibration isolation theory, when the external excitation frequency is much higher than the system's natural frequency, the vibration amplitude transmitted to the vibration-isolated object (i.e., the sensor node) will be attenuated. In this embodiment, the vibration isolation efficiency of this structure in the frequency range above 1 kHz is better than 90%, effectively cutting off the transmission path of frame vibration noise and ensuring that the MEMS microphone unit mainly picks up the sound wave signal propagating through the air medium, thereby improving the signal-to-noise ratio. The calculation of the transmissibility and vibration isolation design of a single-degree-of-freedom damped vibration system are well-known technologies in the field of mechanical engineering and will not be elaborated here.

[0050] Please see the appendix Figure 4 This embodiment elaborates on the sound signal acquisition and numbering process in step S1. Step S1 specifically includes the following sub-steps: S101, Array Clock Driving and Raw Stream Acquisition. The clock management unit inside the sound acquisition module 104 generates a unified driving clock signal and synchronously sends it to each MEMS microphone unit in the first MEMS microphone array 101 to the third MEMS microphone array 103 via the clock bus on the circuit board. In this embodiment, the driving clock frequency... The frequency is set to 3.072 MHz. After receiving the drive clock signal, each MEMS microphone unit oversamples and modulates the external sound pressure, outputting a 1-bit pulse density modulated data stream. The pulse density in this PDM data stream corresponds to the instantaneous amplitude of the analog sound wave signal. The data stream is transmitted back to the input / output interface of the sound acquisition module 104 through the data lines inside each array node.

[0051] S102. Digital Decimation Filtering and Format Conversion. After receiving multiple parallel PDM data streams, the sound acquisition module 104 instantiates a cascaded integrator comb filter for each signal in the FPGA's internal logic. The CIC filter is used to remove high-frequency quantization noise from the PDM signal and perform downsampling processing, converting it into high-linearity pulse code modulation data. The downsampling factor set in this embodiment... The final effective sampling rate of the system is 64. With drive clock frequency and downsampling factor The quantitative relationship between them satisfies the following formula: ; Substituting the aforementioned values ​​into the formula, the effective sampling rate is calculated. The frequency is 48 kHz. After processing by the CIC filter, the original 1-bit data stream is converted into linear PCM audio data with a bit depth of 24 bits. This data format can preserve dynamic range details within the 750 Hz to 20 kHz frequency band. The specific register transfer stage code implementation of the CIC filter and the stage configuration of the integrator and comb are well known in the art and will not be described in detail here.

[0052] S103, Clock Synchronization and Data Frame Encapsulation. To ensure the accuracy of subsequent time-difference-of-arrival (TDOA) positioning algorithms, all acquisition channels must operate on the same time base. The sound acquisition module 104 internally maintains a 32-bit global counter, driven by a high-precision crystal oscillator, used to generate millisecond-level timestamps. The sound transmission and number storage module 105 receives each frame with a length of... After processing PCM data (e.g., 1024 sampling points), data packets are encapsulated.

[0053] The encapsulated data frame structure includes a frame header, metadata segment, and payload data segment. The frame header identifies the start position of the data packet. The metadata segment contains an 8-bit array ID, an 8-bit microphone channel ID, and a 32-bit global timestamp. The array ID uniquely corresponds to the first MEMS microphone array 101, the second MEMS microphone array 102, or the third MEMS microphone array 103; the microphone channel ID uniquely corresponds to a specific unit within each array. The payload data segment is filled with a 24-bit PCM audio sampling sequence for that time period. The encapsulated data frame is sent to the signal processing module 200 via an RS-485 differential bus, realizing the conversion of acoustic physical quantities into a digital information stream with configuration and timing information.

[0054] This embodiment provides a detailed explanation of the audio signal time-frequency filtering and frame-segmentation windowing process in step S2 described above. The signal processing module 200 executes the following sub-steps through its internally integrated digital signal processing unit.

[0055] S201. Digital bandpass filtering. After receiving the pulse-code modulation data stream from the sound transmission and numbering storage module 105, the signal processing module 200 first performs time-domain filtering on it. This embodiment uses a 512th-order finite-length unit impulse response filter. The passband frequency range of this filter is set to 1 kHz to 8 kHz. The basis for this frequency band setting is that the motor roaring noise and low-frequency structural vibration generated during the normal operation of the scraper conveyor are mainly concentrated below 1 kHz, while the fault characteristic sounds of scraper chain crack propagation and tooth surface impact, which are of interest in this invention, are mainly distributed in the mid-to-high frequency range of 1 kHz to 8 kHz. Through this bandpass filter, DC components, power frequency interference, and low-frequency mechanical background noise are effectively filtered out, while high-frequency electronic thermal noise above 8 kHz is suppressed. The specific coefficient design and convolution operation implementation of the FIR filter are well known in the art and will not be described in detail here.

[0056] S202, Signal Overlapping Framing. Due to the non-stationary nature of the scraper conveyor's operation, the signal processing module 200 performs framing processing on the filtered long-time signal sequence to extract local stationary features. Let the filtered discrete-time signal be... It is then divided into several short frames. The length of each frame is... The sampling points are set to 1024. To avoid data discontinuity between adjacent frames and to preserve frame edge information, an overlapping framing strategy is adopted, with the frame shift length set to half the frame length, i.e., 512 sampling points, corresponding to an overlap rate of 50%. This means that the starting point of the mm-th frame data is the mm-th frame in the original signal sequence. One point.

[0057] S203. Time-domain windowing. Directly performing a Fourier transform on the truncated signal frame can lead to spectral leakage, where signal energy leaks from the main lobe to the side lobes, thus masking weak fault characteristic frequencies. Therefore, the signal processing module 200 multiplies each frame of signal by a window function before performing the frequency domain transform. This embodiment uses a Hamming window, which has a low side lobe peak value and can effectively improve spectral resolution.

[0058] Let the first The original signal of the frame is The signal after windowing is The windowing operation satisfies the following formula: ; in, The Hamming window function is expressed mathematically as follows: ; in, This represents the length of the frame, which is 1024. Represents the index of the sampling point within the frame; Pi; The cosine function is used. Through the above calculations, the signal, originally rectangularly truncated in the time domain, is smoothly attenuated to near zero at both ends, resulting in a smooth transition at the beginning and end of the frame signal, generating a windowed signal. It is cached in the memory of the signal processing module 200, waiting to be processed by the subsequent fast Fourier transform.

[0059] Please see the appendix Figure 5 - Appendix Figure 8 This embodiment elaborates on the spectrum solution, power spectral density calculation, and feature extraction processes of steps S3 and S4 described above. The signal processing module 200 executes the following sub-steps.

[0060] S301. Frequency domain transformation and single-frame periodogram calculation. The signal processing module 200 processes the windowed time-domain signal of each frame in the buffer. Perform an arbitrary point Fast Fourier Transform to map the signal from the time domain to the frequency domain. Then, calculate the square of the magnitude of the transformed signal to obtain the first... Periodic diagram of frame signal This corresponds to a specific physical frequency value. This step converts the vibration waveform of a sound wave into a spectrum showing the energy distribution as a function of frequency.

[0061] S302, Welch average power spectral density estimation. Because the periodogram of a single-frame signal is affected by random noise and has a large variance, it cannot be directly used as a stable feature. Signal processing module 200 uses the Welch method to estimate the average power spectral density of continuous signals. The frame periodicity graph is smoothed. In this embodiment, the number of smoothing frames is set. The value is 64. The signal processing module 200 handles the data in the buffer. The frame periodogram is accumulated at the corresponding frequency cell positions and the average value is calculated to obtain the smoothed power spectral density. The calculation process satisfies the following formula: ; in, Indicates the first Power spectral density values ​​at each frequency unit; The total number of frames participating in the averaging; For the first The frame signal is in the first The periodogram values ​​at each frequency unit. Through this step, sudden random disturbances are suppressed, while the persistent periodic fault characteristics of the scraper chain are enhanced in the spectrum.

[0062] S401, Subband Energy Convergence. To reduce feature dimensionality and focus on fault-sensitive frequency bands, signal processing module 200 performs power spectral density... Based on this, and according to the acoustic mechanism of scraper chain failure, five key frequency sub-bands were identified. The center frequencies of these five sub-bands were set at 1.5 kHz, 3 kHz, 4.5 kHz, 6 kHz, and 7.5 kHz, respectively, and each sub-band had a fixed bandwidth (e.g., 500 Hz). The signal processing module 200 calculated the frequency range covered by each sub-band. The summation yields five energy values, which are arranged in ascending order of frequency to form the initial feature vector. .in, Indicates the first The total energy within each sub-band.

[0063] S402, Feature Vector Normalization and Logarithmic Transformation. Due to variations in the load on the scraper conveyor and changes in the distance between the microphone and the sound source, the overall amplitude of the received signal fluctuates. To eliminate the impact of this scale difference on fault identification, the signal processing module 200 normalizes and logarithmically transforms the initial feature vector. L2 norm normalization is performed, and the natural logarithm of the normalized components is taken to generate the final fault feature vector. The transformation process satisfies the following formula: ; in, This is the final fault feature vector output. The initial feature vector; This represents taking the natural logarithm of each element in the vector; Represents the initial feature vector The Euclidean norm (L2 norm) of the feature vector is calculated as the square root of the sum of the squares of all elements in the vector. Through L2 normalization, the direction of the feature vector remains unchanged while its magnitude is standardized, eliminating the influence of signal strength. Logarithmic transformation maps the energy values ​​with a large dynamic range to a linear space, making the feature distribution closer to a Gaussian distribution, which facilitates subsequent similarity calculations. The final generated fault feature vector... As an acoustic fingerprint of the scraper's operating status during that time period, it is sent to the subsequent fault identification unit.

[0064] Figure 5 and Figure 6 It is the sound produced by the scraper movement when the scraper conveyor is working normally. It has periodic pulses, and the frequency spectrum display shows that there are 4k-10k frequency components when there is a collision. Figure 7 and Figure 8 It is the sound emitted by the scraper movement when the scraper conveyor malfunctions. The periodic pulses disappear at a specific moment, and the spectrum frequency display shows no 4k-10k frequency components during the collision.

[0065] This embodiment elaborates on the comparison with the preset voiceprint database and fault diagnosis process in step S5. It includes the following sub-steps: S501. Loading and Initialization of the Fault Voiceprint Library. The non-volatile storage unit of the signal processing module 200 contains a pre-installed standard fault voiceprint library. This library is not a simple collection of audio samples, but rather a set of feature parameters obtained through offline training and cluster analysis. The voiceprint library contains... Typical scraper chain failure modes include chain breakage, severe scraper bending, missing sprocket teeth, and chain skipping. Each failure mode... (in ) by a high-dimensional centroid vector Unique representation. High-dimensional centroid vector With real-time fault feature vector They have the same dimension and are all composed of the normalized logarithmic energy of five key frequency sub-bands. During the system startup initialization phase, the signal processing module 200 loads these centroid vectors into the high-speed computing cache.

[0066] S502, Feature Vector Similarity Calculation. To quantify the degree of matching between the current operating state and known fault modes, the signal processing module 200 calculates real-time fault feature vectors one by one. With each high-dimensional centroid vector in the voiceprint database The cosine similarity between two vectors. Cosine similarity effectively assesses the consistency of two vectors along directions in the feature space, unaffected by the absolute magnitude of the vectors. For the ... Similarity coefficient of class faults The calculation satisfies the following formula: ; in, Represents real-time fault feature vector The Euclidean norm; Indicates the first The Euclidean norm of the centroid vector of the fault class. Signal processing module 200 executes in parallel. The above operation yields a result containing A set of similarity values .

[0067] S503, Optimal Match Fault Type Determination. The signal processing module 200 searches for the maximum value and its corresponding index from the calculated similarity set. If the maximum value is less than the preset background noise threshold, the scraper conveyor is determined to be in normal operation, and the fault handling logic is not triggered. If the maximum value is greater than or equal to the preset minimum judgment threshold... (For example, 0.85), then it is preliminarily determined that the [number]th [event] has occurred. Type of failure.

[0068] S504, Timing Consistency Check. To prevent false alarms caused by occasional impacts from hard objects or electromagnetic interference, the system introduces a timing continuity check mechanism. The signal processing module 200 maintains a queue of status registers, which only checks for continuous... The identification results for all frames (e.g., 5 consecutive frames, approximately 100 milliseconds) are of the same fault type. And each frame A fault is only confirmed when all thresholds are exceeded.

[0069] S505, Fault Confidence Classification and Coding. After confirming the occurrence of a fault, the signal processing module 200... The magnitude of the value is used to assess the fault level. If (in If a high confidence threshold (e.g., 0.95) is set, the system generates a status code requiring verification, indicating a suspected fault; if The system generates a high-confidence status code, indicating the existence of a definite fault. Finally, the signal processing module 200 indexes the determined fault type. The confidence status code is combined to generate a fault type code, which is then ready to proceed to the subsequent location and feedback process.

[0070] This embodiment elaborates on the multi-channel signal fault location calculation process in step S6 above. After determining that there is a fault in the scraper chain, in order to provide maintenance personnel with accurate maintenance guidance, the signal processing module 200 uses the signal time difference between the first MEMS microphone array 101, the second MEMS microphone array 102, and the third MEMS microphone array 103 distributed at different positions on the scraper conveyor to calculate the three-dimensional spatial coordinates of the fault. This process specifically includes the following sub-steps: S601. Multi-channel signal synchronous extraction and pairing. The signal processing module 200 retrieves the original pulse code modulation data within a preset time window before and after the fault alarm trigger time from its internal high-speed cache. The signal processing module 200 selects microphone signals from different array nodes to construct cross-correlation pairs. Typically, any microphone channel in the first MEMS microphone array 101 is selected as the reference signal. The microphone channel in either the second MEMS microphone array 102 or the third MEMS microphone array 103 is selected as the comparison signal. Since all acquisition nodes have been synchronized using a unified global clock in step S1, therefore and They are strictly aligned on the time axis, and any phase difference is caused only by the difference in the distance the sound waves travel.

[0071] S602. Generalized Cross-Correlation Delay Estimation. To calculate the time difference between sound source propagation to different microphone arrays, signal processing module 200 performs a generalized cross-correlation operation on selected signal pairs. Although ambient noise affects signal waveforms, acoustic events caused by faults (such as tooth impact) exhibit correlated waveform characteristics in signals received at different locations. Signal processing module 200 calculates the cross-correlation function of the two signals. And search for the peak location of the function to determine the time delay. The calculation process satisfies the following formula: ; in, Indicates the sound wave reaches the th The microphone relative to reaching the first Time delay estimates for each microphone; This represents the sliding time variable in the related calculations; and These represent the time-domain signals received by the two microphones, respectively. This represents the value of the variable that maximizes the function; integration is achieved through discrete summation in digital systems. This step converts the physical sound path difference into a quantifiable microsecond-level time difference value.

[0072] S603. Construction and coordinate calculation of hyperboloid equations. The signal processing module 200, based on the pre-calibrated and stored geometric coordinates of each MEMS microphone array, combines the calculated time delay... Establish a set of equations for sound source localization. Let the first equation be... The spatial coordinates of the microphones are , No. The spatial coordinates of the microphones are The coordinates of the sound source (i.e., the fault point) to be determined are: According to the distance difference principle, the sound source lies on a hyperboloid of revolution with the two microphone positions as foci and the distance difference between them as constant. This geometric relationship satisfies the following equation: ; in, The speed of sound in the air medium underground is represented by a value of 340 m / s. The signal processing module 200 uses at least three different microphone pairs to establish a system of simultaneous equations, and solves these nonlinear equations using the least squares method or Taylor series expansion iterative method to obtain the three-dimensional spatial coordinates of the sound source. For linearly distributed scraper conveyors, the longitudinal component of this coordinate system (usually along the roadway direction) directly corresponds to the specific number of chute sections or the position of the frame.

[0073] S604. Encapsulation of Positioning Results. The signal processing module 200 associates and packages the calculated three-dimensional spatial coordinates with the fault type code and fault occurrence timestamp generated in the previous steps to form a complete fault detection report data packet. This data packet is sent to the subsequent feedback decision layer through the downhole industrial ring network interface to guide maintenance work in physical space.

[0074] Please see the appendix Figure 9 This embodiment elaborates on the output quality evaluation result of step S7 and the alarm triggering and data transmission process in the closed-loop feedback. Specifically, it includes the following sub-steps: S701, Edge-side Alarm Logic Trigger. The signal processing module 200 has a digital output interface, which connects to an intrinsically safe audible and visual alarm installed in the scraper conveyor tunnel. When the signal processing module 200 determines that the fault confidence status code is high confidence, it immediately outputs a high-level control signal through the digital output interface, driving the audible and visual alarm to emit a red flashing light and intermittent buzzing sound, prompting on-site personnel to take precautions or stop the machine for inspection. When the fault confidence status code requires verification, the signal processing module 200 outputs a pulse control signal, driving the audible and visual alarm to emit a yellow warning light, indicating that attention is needed. This hierarchical alarm mechanism is directly based on the real-time judgment results of the edge computing node, without relying on instructions from the ground server, ensuring the real-time response to sudden faults.

[0075] S702, Communication Protocol Encapsulation. The signal processing module 200 encapsulates the detection results into data frames conforming to industrial communication standards. This embodiment uses a custom application layer message format based on the TCP / IP protocol suite. The payload fields of the data frame include: a unique device identifier, a millisecond-level Unix timestamp at the time of the fault, and a fault type index code. Fault confidence similarity value and the three-dimensional spatial coordinates of the fault source. To ensure data integrity during transmission, the signal processing module 200 calculates a cyclic redundancy check (CRC) code for the payload field and appends the check code to the end of the data frame.

[0076] S703, Data Upload and Routing. The signal processing module 200 connects to the underground industrial ring network switch via its integrated industrial Ethernet interface. Based on the pre-configured network routing table, the signal processing module 200 establishes Socket connections with the monitoring host 301 and the remote server 303. The encapsulated data frames are concurrently sent via the underground fiber optic ring network to the monitoring host 301 located in the roadway control room and the remote server 303 located in the surface data center. Upon receiving the data frames, the monitoring host 301 parses the three-dimensional spatial coordinates within them. This fault location is mapped onto the digital twin model of the scraper conveyor in the 3D visualization interface, and then highlighted and flashed to visually display the fault location for the control room operator's decision-making reference. The specific implementation of the TCP / IP protocol stack and network socket programming are well-known technologies and will not be elaborated upon here.

[0077] Please see the appendix Figure 10 To verify the accuracy of the automatic recognition results from the signal processing module 200 and to accumulate labeled data for subsequent model optimization, the system introduced a manual review mechanism via mobile terminal 302. This process specifically includes the following steps: S704, Maintenance Task Push and Navigation. After the message middleware service built into the remote server 303 detects newly uploaded fault alarm data, it sends a push notification to the mobile terminal 302 held by the on-duty maintenance personnel via the underground wireless LAN, based on preset scheduling logic. The push notification uses a lightweight message queue telemetry transmission protocol. After receiving the notification, the application on the mobile terminal 302 parses the three-dimensional spatial coordinates in the message. In conjunction with the installation location diagram of the scraper conveyor, the specific frame number or chute section number of the fault point is displayed graphically on the screen, guiding maintenance personnel to the designated location.

[0078] S705. On-site Status Confirmation and Multimodal Data Acquisition. After arriving at the designated location, maintenance personnel visually inspect the actual condition of the scraper chain. The mobile terminal 302 provides a standardized verification interface, where maintenance personnel select either fault confirmation or false alarm options. If a fault is confirmed, maintenance personnel must further use the mobile terminal 302 to take high-resolution photos of the scraper chain on-site and record a near-field audio segment. The recording duration of this near-field audio is set to 10 to 30 seconds. Since the mobile terminal 302 is stationary at close range, the signal-to-noise ratio of its recorded audio signal is higher than that of a fixedly installed MEMS microphone array; this audio will serve as a high-confidence reference sample. The development of the APP interface for the mobile terminal 302 and the retrieval of multimedia data are well-known technologies in this field and will not be elaborated upon here.

[0079] S706. Verification Data Transmission and Tag Association. The mobile terminal 302 packages the confirmation result input by the maintenance personnel, the captured photos, and the recorded near-field audio into a verification data packet, which is then transmitted back to the remote server 303 via the wireless network. Upon receiving the verification data packet, the remote server 303 retrieves the original array sound data segment that triggered the alarm from the database based on the timestamp of the fault occurrence and the device ID. The remote server 303 associates the verification result as a tag with the original array sound data. If the verification result is a false alarm, the data is marked as a negative sample; if the verification result is a fault confirmation, the data is marked as a positive sample, and the near-field audio recorded by the mobile terminal 302 is stored as an auxiliary feature in the sample library. This step completes the transformation of unsupervised raw data into a supervised training dataset, providing a data foundation for the iteration of the algorithm model.

[0080] Please see the appendix Figure 11 The remote server 303 dynamically updates the fault identification algorithm model using manually verified data and sends the updated parameters to the downhole signal processing module 200 via over-the-air (OTA) download technology. This process specifically includes the following sub-steps: S707, Incremental update of feature centroid vector. Remote server 303 periodically (e.g., after each shift) executes a batch job. The server retrieves from the database recently manually reviewed and marked as faulty, belonging to the [number missing] category. A set of positive sample data for the fault class. Suppose this batch contains... A valid fault feature vector (in Remote server 303 calculates the average feature vector of this batch of samples, and uses a weighted moving average algorithm to calculate the centroid vector of the corresponding kk-th type of fault in the voiceprint database. Perform an update. The update calculation satisfies the following formula: ; in, Indicates the updated number Fault centroid vector; This represents the old centroid vector before the update; This represents the update factor (or learning rate value), which is typically set between 0.01 and 0.1. Learning rate value The setting of this value determines the model's sensitivity to new data: a smaller value allows the model to retain more historical features and has higher stability; a larger value allows the model to adapt more quickly to the current wear and tear of the equipment. Using this formula, the system can fine-tune the voiceprint features for specific fault modes without retraining the entire model.

[0081] S708, Upgrade Package Generation and Verification. After completing the centroid vector update for all fault categories, remote server 303 will send the updated centroid vector set. And the new judgment threshold adjusted based on the false alarm rate. Serialize to a binary configuration file. The remote server assigns a new version number to the configuration file and calculates the MD5 digest value of the file as an integrity checksum.

[0082] S709. Parameter Distribution and Integrity Verification. The remote server 303 establishes a file transfer connection with the signal processing module 200 via the underground industrial ring network. The binary configuration file and its corresponding MD5 digest value are sent to the temporary storage area of ​​the signal processing module 200. Upon receiving the file, the signal processing module 200 immediately calculates the MD5 value of the locally received file and compares it with the received digest value. If they match, the transmission process is error-free, and the loading process begins; if they do not match, the file is discarded and a retransmission is requested, thus preventing parameter errors due to network errors.

[0083] S710, Double-Buffered Hot Loading. To ensure the continuity of online monitoring and avoid system downtime due to parameter updates, the signal processing module 200 employs a double-buffering mechanism to load new models. The signal processing module 200 allocates two independent storage spaces in its memory: a running area and a loading area. The currently executing algorithm task reads the voiceprint library parameters from the running area. The new parameters downloaded in step S709 are written to the loading area. After writing is complete and verification passes, the signal processing module 200, during the interval between processing the current frame of audio data, switches the memory pointer through atomic operations, marking the loading area as active and the original running area as idle. When the next frame of audio data arrives, the algorithm directly calls the newly loaded voiceprint library parameters for calculation. This mechanism achieves seamless switching at the millisecond level, ensuring zero-interruption operation of the scraper conveyor monitoring system. The double-buffering technology and memory pointer operations are well-known technologies in this field and will not be elaborated upon here.

[0084] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for online non-destructive acoustic testing of scraper conveyors, characterized in that, Includes the following steps: S1. The sound acquisition layer converts the acquired signal into pulse code modulation data and uploads it to the signal processing layer. S2. The signal processing layer performs bandpass filtering, framing, and windowing on the pulse code modulation data to obtain a windowed signal. S3. Calculate the power spectral density of the windowed signal using the Welch method; S4. Normalize and logarithmically transform the subband energy integral of the power spectral density to generate a fault feature vector. S5. Calculate the cosine similarity between the fault feature vector and the fault voiceprint database to determine the fault. S6. Solve the three-dimensional spatial coordinates of the fault using multi-channel time delay and hyperboloid intersection method; S7. Output the three-dimensional spatial coordinates and issue an alarm, and update the fault voiceprint database based on the manual review data from the feedback decision layer.

2. The method for online non-destructive acoustic testing of a scraper conveyor according to claim 1, characterized in that, In step S1, the process of converting the data into pulse code modulation data includes: It receives multiple pulse density modulated signals, uses a cascaded integrator comb filter algorithm for downsampling and high-frequency quantization noise removal, and converts a 1-bit data stream into 24-bit linear audio data. The effective sampling rate of the downsampling is calculated by dividing the driving clock frequency by the downsampling factor.

3. The method for online non-destructive acoustic testing of a scraper conveyor according to claim 1, characterized in that, In step S2, the bandpass filter uses a finite-length unit impulse response filter with a frequency range of 1 kHz to 8 kHz. The windowing process utilizes the Hamming window function formula to calculate each frame of signal, thereby suppressing spectral leakage.

4. The method for online non-destructive acoustic testing of a scraper conveyor according to claim 1, characterized in that, In step S4, the specific steps for generating the fault feature vector are as follows: The power spectral density is obtained by averaging the periodograms of multiple consecutive frames using the Welch averaging method. The summation algorithm is used to calculate the total energy of the power spectral density within the five key frequency sub-bands to form the initial feature vector; The initial eigenvectors are normalized using the L2 norm formula to eliminate signal strength differences; The final fault feature vector is generated by calculating the natural logarithm transformation method for each element in the normalized vector.

5. The method for online non-destructive acoustic testing of a scraper conveyor according to claim 1, characterized in that, Step S5 also includes a fault confidence grading step: If the cosine similarity is greater than or equal to the minimum judgment threshold and less than the high confidence threshold, a status code requiring review is generated, and the audible and visual alarm emits a yellow warning light. If the cosine similarity is greater than or equal to the high confidence threshold, a high confidence status code is generated, driving the audible and visual alarm to emit a red flashing light and an intermittent buzzing sound.

6. The method for online non-destructive acoustic testing of a scraper conveyor according to claim 1, characterized in that, Step S6 specifically includes: We select microphone signals from different array nodes to construct cross-correlation pairs, use the generalized cross-correlation algorithm to calculate the cross-correlation function, search for the peak position of the cross-correlation function, and determine the delay of sound waves arriving at different arrays. Based on the principle of distance difference, the system of equations for a hyperboloid with the microphone position as the focus is constructed. The least squares method and the iterative method are used to solve the system of equations for the hyperboloid and calculate the three-dimensional spatial coordinates of the sound source.

7. The method for online non-destructive acoustic testing of a scraper conveyor according to claim 1, characterized in that, In step S7, the process of obtaining the manually reviewed data includes: The mobile terminal receives a push notification and displays the rack number of the fault location; Receive near-field audio recordings and photos taken by maintenance personnel; If the verification result confirms the fault, the original array sound data is marked as a positive sample, and the near-field audio is stored in the sample library as an auxiliary feature.

8. The method for online non-destructive acoustic testing of a scraper conveyor according to claim 7, characterized in that, In step S7, the process of updating the fault voiceprint database includes: Extract the dataset that has been manually reviewed and marked as positive samples, and calculate the average feature vector of the current batch of samples using the arithmetic mean algorithm. Using the weighted moving average formula, based on the configured update factor, calculate the weighted sum of the old centroid vector and the average feature vector of the current batch to obtain the updated centroid vector.

9. The method for online non-destructive acoustic testing of a scraper conveyor according to claim 1, characterized in that, In step S7, the process of sending parameters to the signal processing layer adopts a double-buffered hot-loading mechanism: The signal processing layer receives the configuration file and uses the MD5 digest algorithm to calculate the MD5 value for integrity verification. After successful verification, the parameters are written to the loading area in memory; During the interval between processing the current frame of data, the memory pointer is switched using atomic operations to mark the loading area as active, so that the next frame of data can call the updated parameters.

10. A scraper conveyor scraper online acoustic non-destructive testing system applying the online acoustic non-destructive testing method for scraper conveyors according to any one of claims 1-9, characterized in that, include: The sound acquisition layer is arranged longitudinally along the body of the scraper conveyor, including the first, second and third MEMS microphone arrays respectively set in the head, tail and middle areas of the body, as well as the sound acquisition module and the sound transmission and numbering storage module. The signal processing layer includes a signal processing module connected to the sound acquisition layer via a bus, which is used to perform time-frequency analysis, fault detection, and location calculation. Feedback to the decision-making level includes a monitoring host, mobile terminal, and remote server connected to the signal processing module via an underground industrial ring network; The MEMS microphone array adopts an array-type sensor node structure, which includes a cast aluminum explosion-proof housing. The front end of the housing is covered with a thin-walled metal cover to form a zero-light-window sealed structure. The housing contains multiple microelectromechanical system microphone units arranged in a circular pattern; The housing is fixed by a composite vibration damping mounting mechanism, which includes a silicone damping seat with a honeycomb-shaped vibration damping hole array, and a metal limiting sleeve is embedded in the damping seat.

Images