A high-precision bearing steel ball internal micro-defect ultrasonic detection method and system

Abstract

This application relates to the field of bearing inspection technology, specifically a high-precision ultrasonic detection method and system for micro-defects inside bearing steel balls. The method includes: acquiring ultrasonic reflection signals from the steel ball at its full depth and digitizing them into a time-domain amplitude sequence; constructing a reflection path tracing logic based on the steel ball's geometric parameters and internal sound velocity to generate a time-series distribution matrix of theoretical reverberation nodes; extracting candidate echo envelopes by dynamically warping and aligning the boundary reverberation components; inputting this envelope into a pulse neural network model that integrates an adaptive threshold membrane potential mechanism to encode time-series pulse features and suppress high-frequency noise; decoding the pulse features to determine the absolute acoustic time of flight; calculating the precise spatial coordinates of the micro-defects by combining the incident angle; and outputting the results. This technical solution effectively eliminates the interference of multiple reverberations at the spherical boundary, achieving high-precision positioning and detection of micro-defects inside bearing steel balls.

Description

Technical Field

[0001] This invention relates to the field of bearing testing technology, specifically to a high-precision ultrasonic testing method and system for micro-defects inside bearing steel balls. Background Technology

[0002] Ultrasonic nondestructive testing technology is widely used for internal quality assessment of metal parts. However, when performing full-depth ultrasonic testing of internal micro-defects in high-precision bearing steel balls, the problem of signal aliasing and masking arises.

[0003] When acquiring full-depth ultrasonic reflection signals from bearing steel balls, the propagation of ultrasonic waves within the enclosed sphere generates dense, multiple reverberant echoes from the spherical boundaries. In the generated time-domain amplitude sequence, these high-intensity boundary reverberant components intertwine and overlap with extremely weak internal micro-defect echoes in the time domain. Existing signal processing methods struggle to accurately isolate these nonlinear boundary reverberant components from complex time-domain signals containing residual high-frequency electrical noise, resulting in the true echo envelope of the internal micro-defects being completely masked by the multiple spherical boundary reverberations. This deep masking of micro-defect characteristics by multiple spherical boundary reverberations prevents the detection system from acquiring a pure internal micro-defect echo pulse sequence, thus hindering the accurate resolution of the micro-defect's absolute acoustic time of flight and ultimately preventing the calculation and output of the micro-defect's precise spatial coordinates within the steel ball.

[0004] In view of this, this application proposes a high-precision ultrasonic testing method and system for micro-defects inside bearing steel balls. Summary of the Invention

[0005] To achieve the above objectives, this application provides a high-precision ultrasonic testing method and system for internal micro-defects in bearing steel balls, the specific technical solution of which is as follows:

[0006] A high-precision ultrasonic testing method for internal micro-defects in bearing steel balls includes:

[0007] Acquire full-depth ultrasonic reflection signals of bearing steel balls and record the corresponding incident spatial angles. Digitize the full-depth ultrasonic reflection signals into a time-domain amplitude sequence containing internal defect echoes and multiple reverberant echoes from the spherical boundary, and mark the time axis coordinates.

[0008] Based on the geometric diameter parameters and internal sound velocity of the steel ball, a path tracing logic for multiple reflections at the spherical boundary is constructed to calculate the transit time of the ultrasonic wave propagating back and forth within the spherical boundary and generate the temporal distribution matrix of the theoretical reverberation nodes.

[0009] The temporal distribution matrix of the theoretical reverberation node is dynamically time-warped and aligned with the temporal amplitude sequence. The boundary reverberation component in the temporal amplitude sequence is removed, and the candidate echo envelope containing internal micro-defect features is extracted.

[0010] The candidate echo envelope is input into a pulse neural network model that integrates an adaptive threshold membrane potential mechanism to encode the temporal pulse characteristics of the candidate echo envelope and suppress residual high-frequency electrical noise to obtain a pure internal micro-defect echo pulse sequence.

[0011] Decode the timing pulse characteristics to determine the absolute acoustic time of flight of the internal micro-defects, combine the incident spatial angle of the ultrasonic probe to calculate the precise spatial coordinates of the micro-defects, and output the ultrasonic detection results of the internal micro-defects of the bearing steel ball.

[0012] Preferably, the acquisition of the full-depth ultrasonic reflection signal of the bearing steel ball includes:

[0013] A focused ultrasonic probe is used to apply broadband pulse excitation to the bearing steel ball in a liquid immersion coupling manner. The focal area of ​​the focused ultrasonic probe is aligned with the geometric center of the steel ball to cover the full depth of the sound field inside the ball.

[0014] The focused ultrasound probe is controlled to perform multi-directional point-by-point rotation scanning along the outer surface of the steel ball according to a preset spatial angle step distance. At each scanning direction, the full-time reflection signal including the front surface echo, the internal scattered echo, and the bottom reflection echo of the sphere is received. The reflection signals from all directions are converged to form the full-depth ultrasound reflection signal.

[0015] Preferably, the step of digitizing the full-depth ultrasonic reflection signal into a time-domain amplitude sequence containing internal defect echoes and multiple spherical boundary reverberation echoes, and marking the time axis coordinates, includes:

[0016] The full-depth ultrasonic reflection signal is sampled by high-speed analog-to-digital conversion, and the continuous analog reflection signal is discretized into a digital amplitude sequence with equal intervals.

[0017] Using the trigger moment of the ultrasonic probe's emitted pulse as the zero point of the time reference, and assigning absolute time coordinates to each sampling point in the digital amplitude sequence according to the sampling interval, a time-domain amplitude sequence with complete time axis marking is generated. The time-domain amplitude sequence contains the superposition components of internal defect echoes and multiple spherical boundary reverberation echoes.

[0018] Preferably, the logic for constructing the spherical boundary multiple reflection path tracing includes:

[0019] Obtain the nominal geometric diameter parameters of the bearing steel balls and the longitudinal wave sound velocity parameters of the steel ball material;

[0020] A geometric model for ultrasonic wave propagation under spherical boundary conditions is established. In the geometric model, a complete ray tracing path is defined for the ultrasonic wave after it enters the sphere from the incident point and is reflected successively by the inner wall of the sphere.

[0021] The spatial geometric relationship between the incident angle and the reflection angle for each reflection is determined based on the curvature of the sphere. The acoustic path length between each two adjacent reflections of the ultrasonic wave within the sphere is calculated, and a path propagation sequence covering multiple reflection orders is constructed.

[0022] Preferably, the calculation of the transit time of the ultrasonic wave propagating back and forth within the spherical boundary includes:

[0023] Based on the path lengths of each order of reflection in the constructed path propagation sequence, and combined with the longitudinal wave velocity parameters of the steel ball material, the path lengths of each order of reflection are converted into corresponding single transit time components.

[0024] Starting from the incident ultrasonic wave on the front surface of the sphere, the cumulative sound path time corresponding to the reflection from the inside of the sphere to the boundary of each order of the sphere and back to the probe receiving surface is successively superimposed to obtain the cumulative transit time value of the echo reflected from the boundary of each order of the sphere to the probe receiving surface.

[0025] Preferably, the temporal distribution matrix of the generated theoretical reverberation nodes includes:

[0026] The cumulative transit time values ​​corresponding to the reflected echoes from the spherical boundary of each order are arranged in order of reflection from low to high to form a one-dimensional theoretical reverberation time node sequence.

[0027] For each theoretical reverberation time node, its corresponding reflection order number and estimated echo amplitude attenuation weight are labeled. The reverberation time node, reflection order number and amplitude attenuation weight are combined to construct a multi-dimensional time-series distribution matrix. Each row in the time-series distribution matrix corresponds to a reverberation event of a specific reflection order and its time-domain characteristic parameters.

[0028] Preferably, the step of dynamically time-warping and aligning the temporal distribution matrix of the theoretical reverberation nodes with the time-domain amplitude sequence includes:

[0029] Using each reverberation time node in the temporal distribution matrix of the theoretical reverberation nodes as a reference template sequence, and the local extreme points in the measured time-domain amplitude sequence as the feature sequence to be matched, a nonlinear elastic time mapping path is established between the reference template sequence and the feature sequence to be matched through a dynamic time warping algorithm. The optimal alignment path that minimizes the cumulative distance cost is searched node by node to complete the one-to-one time alignment between the theoretical reverberation time nodes and the corresponding boundary reverberation echo peaks in the measured time-domain amplitude sequence.

[0030] Preferably, the stripping of the boundary reverberation component from the time-domain amplitude sequence includes:

[0031] Based on the optimal alignment path obtained by dynamic time warping alignment, the time window position and duration range of each order of spherical boundary reverberant echo are located in the measured time-domain amplitude sequence.

[0032] Within each reverberation time window, a local reverberation signal estimation template is constructed based on the amplitude attenuation weights of the corresponding order in the time-series distribution matrix. The signal components of the measured time-domain amplitude sequence within each reverberation time window are adaptively canceled with the corresponding local reverberation signal estimation template, and the spherical boundary reverberation components are stripped off step by step.

[0033] Preferably, the extraction of candidate echo envelopes containing internal micro-defect features includes:

[0034] In the residual time-domain signal after stripping the boundary reverberation component, a signal energy detection threshold is set to identify effective echo segments in the residual time-domain signal that exceed the signal energy detection threshold.

[0035] The amplitude envelope curves of each identified effective echo segment are extracted. Isolated non-defect random interferences are eliminated by envelope peak clustering analysis, and effective echo segments suspected to be internal micro-defect scattering sources are screened out.

[0036] The selected valid echo segments, along with their time coordinate information, are extracted to form a candidate echo envelope containing internal micro-defect features.

[0037] Preferably, the spiking neural network model includes a multi-layer network architecture consisting of an input coding layer, a spiking processing hidden layer, and an output decoding layer;

[0038] The input coding layer converts the amplitude timing signal of the candidate echo envelope into a pulse sequence and inputs it into the pulse processing hidden layer;

[0039] Each spiking neuron in the hidden layer of the pulse processing adopts a leakage integral firing model. Each spiking neuron maintains a membrane potential state variable that evolves dynamically over time. When the accumulated membrane potential state variable exceeds the firing threshold, the spiking neuron generates an output pulse and resets the membrane potential to the baseline level.

[0040] Preferably, the adaptive threshold membrane potential mechanism includes:

[0041] A dynamically adjustable firing threshold is set for each spiking neuron in the pulse processing hidden layer. The firing threshold is adaptively updated based on the historical pulse firing frequency of the spiking neuron in the recent time window.

[0042] When the cumulative firing count of a spiking neuron within a preset statistical time window exceeds the upper frequency limit, the firing threshold of the spiking neuron is adjusted upward to raise its triggering condition; when the cumulative firing count is lower than the lower frequency limit, the firing threshold of the spiking neuron is adjusted downward to lower its triggering condition, thereby achieving dynamic adaptive adjustment of the firing threshold of each spiking neuron.

[0043] Preferably, encoding the temporal pulse characteristics of the candidate echo envelope and suppressing residual high-frequency electrical noise includes:

[0044] The amplitude variation of the candidate echo envelope is converted into pulse firing time encoding through the input coding layer, and the time-domain aggregation and filtering of the input pulse sequence is performed by utilizing the membrane potential integral characteristics of each spiking neuron in the pulse processing hidden layer.

[0045] Based on the membrane potential leakage mechanism, the irregular short-interval pulses corresponding to the residual high-frequency electrical noise are integrally attenuated to prevent them from accumulating to the firing threshold; the effective time-series pulse sequence output by the pulse-processed hidden layer is determined as the encoded time-series pulse feature.

[0046] Preferably, the decoding of timing pulse features to determine the absolute acoustic time of flight of internal micro-defects includes:

[0047] The pulse timing features output by the spiking neural network model are analyzed to identify the pulse clusters corresponding to each micro-defect echo and extract the start emission time of each pulse cluster.

[0048] The initial emission time of each pulse cluster is mapped back to the time axis coordinate system of the original time domain amplitude sequence. The complete time interval from the moment the ultrasonic probe emits a pulse to the moment the micro-defect scattering echo reaches the probe receiving surface is calculated. The complete time interval is determined as the absolute acoustic flight time of the ultrasonic wave from the probe through the coupling medium into the steel ball to the micro-defect location and back.

[0049] Preferably, the step of calculating the precise spatial coordinates of the micro-defects by combining the incident spatial angle of the ultrasonic probe and outputting the ultrasonic detection results of the micro-defects inside the bearing steel ball includes:

[0050] The arrival time of the front surface echo is extracted from the full-depth ultrasonic reflection signal to determine the one-way propagation time of the ultrasonic wave in the coupling medium;

[0051] Based on the determined absolute acoustic flight time, the two-way propagation time of the ultrasonic wave in the coupling medium is subtracted, and the remaining flight time is converted into the radial depth distance of the micro-defect relative to the incident surface by combining the sound velocity inside the steel ball.

[0052] Obtain the incident spatial angle parameters of the ultrasonic probe at the current scanning orientation, and combine them with the radial depth distance to locate the micro-defect to the precise spatial coordinates of the three-dimensional coordinate system inside the steel ball;

[0053] The spatial coordinates and echo characteristics of micro-defects detected from each scanning orientation are summarized to generate and output the ultrasonic detection results of micro-defects inside the bearing steel ball.

[0054] A high-precision ultrasonic testing system for micro-defects inside bearing steel balls, used to implement the aforementioned high-precision ultrasonic testing method for micro-defects inside bearing steel balls, includes: an ultrasonic signal acquisition module, a reverberation matrix generation module, an echo envelope extraction module, a pulse feature encoding module, and a defect location output module.

[0055] The ultrasonic signal acquisition module acquires the full-depth ultrasonic reflection signal of the bearing steel ball and records the corresponding incident spatial angle. It digitizes the full-depth ultrasonic reflection signal into a time-domain amplitude sequence containing internal defect echoes and multiple reverberation echoes from the spherical boundary, and marks the time axis coordinates.

[0056] The reverberation matrix generation module, based on the geometric diameter parameters and internal sound velocity of the steel ball, constructs a path tracing logic for multiple reflections at the spherical boundary, calculates the transit time of the ultrasonic wave propagating back and forth within the spherical boundary, and generates a temporal distribution matrix of theoretical reverberation nodes.

[0057] The echo envelope extraction module dynamically aligns the temporal distribution matrix of the theoretical reverberation node with the time-domain amplitude sequence, removes the boundary reverberation components from the time-domain amplitude sequence, and extracts candidate echo envelopes containing internal micro-defect features.

[0058] The pulse feature encoding module inputs the candidate echo envelope into the pulse neural network model that integrates the adaptive threshold membrane potential mechanism, encodes the temporal pulse features of the candidate echo envelope, and suppresses residual high-frequency electrical noise to obtain a pure internal micro-defect echo pulse sequence.

[0059] The defect location output module decodes the timing pulse characteristics to determine the absolute acoustic time of flight of the internal micro-defect, calculates the precise spatial coordinates of the micro-defect by combining the incident spatial angle of the ultrasonic probe, and outputs the ultrasonic detection results of the internal micro-defect of the bearing steel ball.

[0060] The beneficial effects of this application are as follows: This application, through full-depth ultrasonic signal acquisition and high-precision digitization, completely preserves the echo and boundary reverberation information of minute defects. Combined with precise marking of spatial angles and time axes, it lays a reliable data foundation for subsequent signal decoupling and three-dimensional positioning, effectively avoiding the loss of early weak defect signals.

[0061] This application combines the geometric and acoustic parameters of a steel sphere to construct a path tracing logic for multiple reflections at the spherical boundary, accurately calculating the transit time of ultrasonic waves. The generated theoretical reverberation time-series distribution matrix provides a scientific basis for subsequent accurate removal of spherical reverberation interference, significantly improving the accuracy of background noise modeling.

[0062] This application introduces a dynamic time warping algorithm to precisely align the theoretical reverberation matrix with the measured sequence, effectively overcoming time offset errors in actual detection. This method accurately isolates the reverberation component at the spherical boundary and successfully extracts the candidate echo envelope of micro-defects, significantly improving the signal-to-noise ratio of the defect signal.

[0063] This application employs a pulse neural network with fused adaptive thresholds to pulse-code candidate echoes. This model can intelligently adapt to signal attenuation, effectively suppress residual high-frequency electrical noise, and obtain highly pure micro-defect echo pulse sequences, significantly enhancing the system's sensitivity to weak defects and its anti-interference capability.

[0064] This application precisely pinpoints the absolute time of flight of micro-defects by decoding pure pulse characteristics. Combined with the probe's incident angle, it achieves accurate three-dimensional coordinate localization of the micro-defects. This step overcomes the limitation of traditional detection methods that can only perform qualitative analysis, providing reliable quantitative data support for the quality rating and precise rejection of bearing steel balls.

[0065] This application effectively addresses the industry pain point of difficulty in detecting micro-defects in high-end bearing steel balls. High-precision three-dimensional non-destructive positioning significantly reduces the risk of defective products entering production and substantially improves the fatigue life and reliability of core bearings. Attached Figure Description

[0066] Figure 1 A flowchart of a high-precision ultrasonic testing method for internal micro-defects in bearing steel balls is provided in this application;

[0067] Figure 2 This is a schematic diagram of a liquid immersion coupled multi-directional ultrasonic scanning scenario provided in this application;

[0068] Figure 3 The full-depth ultrasonic reflection signal and digital schematic diagram provided for this application;

[0069] Figure 4 A schematic diagram of multiple reflection paths of a spherical boundary provided for this application;

[0070] Figure 5 A schematic diagram of boundary reverberation component stripping and candidate echo envelope extraction provided for this application;

[0071] Figure 6 This application provides a structural diagram of a high-precision ultrasonic testing system for micro-defects inside bearing steel balls. Detailed Implementation

[0072] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the specific embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0073] Many specific details are set forth in the following description in order to provide a full understanding of this application. However, this application may also be implemented in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of this application. Therefore, this application is not limited to the specific embodiments disclosed below.

[0074] Secondly, the term "an embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of this application. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single embodiment or an embodiment selectively excluded from other embodiments.

[0075] Example 1

[0076] Reference Figures 1 to 5 This is the first embodiment of the present application, such as Figure 1 As shown, a high-precision ultrasonic testing method for micro-defects inside bearing steel balls is provided.

[0077] Step 1: Acquire the full-depth ultrasonic reflection signal of the bearing steel ball and record the corresponding incident spatial angle. Digitize the full-depth ultrasonic reflection signal into a time-domain amplitude sequence containing internal defect echoes and multiple reverberation echoes from the spherical boundary, and mark the time axis coordinates.

[0078] In the initial stage of the ultrasonic testing process for micro-defects inside bearing steel balls, it is necessary to acquire ultrasonic reflection signals from all directions and depths of the bearing steel balls under inspection, and convert the acquired analog reflection signals into digital time-domain sequences that can be used for subsequent analysis and processing. The entire signal acquisition and digitization process includes the selection of ultrasonic excitation mode, the implementation of acoustic beam coupling and focusing strategies, the planning and execution of multi-directional scanning paths, and high-speed digital sampling and time axis marking.

[0079] To achieve effective ultrasonic excitation and acoustic beam coupling, a focusing ultrasonic probe is used as an integrated transmitting and receiving sensing unit, applying broadband pulse excitation to the bearing steel ball via liquid immersion coupling. Specifically, the bearing steel ball under test is completely immersed in a liquid coupling medium, typically deionized water that has undergone degassing treatment, to ensure stable and uniform acoustic characteristics during ultrasonic wave propagation and to avoid reduced signal-to-noise ratio due to sound wave scattering caused by microbubbles or impurities in the medium. The operating frequency of the focusing ultrasonic probe is determined based on the diameter of the bearing steel ball under test and the minimum target size of the micro-defects to be detected. The frequency selection must balance the penetration ability of ultrasonic waves in the steel ball material with the scattering sensitivity to small defects. The probe's focusing method employs a spherical piezoelectric crystal self-focusing or acoustic lens focusing structure, aligning the acoustic beam focal area with the geometric center of the steel ball, ensuring sufficient sound field coverage intensity of ultrasonic energy throughout the entire depth range from the incident surface to the bottom of the sphere. The broadband pulse excitation signal is generated by an ultrasonic pulse generator. The bandwidth of the pulse signal should cover the effective frequency range on both sides of the nominal center frequency of the probe to ensure that the emitted sound beam has a short time-domain pulse width, thereby providing sufficient axial resolution for subsequent signal processing.

[0080] like Figure 2 The diagram illustrates a liquid-immersion coupled multi-directional ultrasonic scanning scenario. This scenario comprises a liquid immersion tank, a focusing ultrasonic probe, and the bearing steel ball under inspection. The bearing steel ball is completely immersed in deionized water that has undergone degassing treatment. The liquid coupling medium fills the propagation space between the probe's emitting surface and the steel ball's incident surface, ensuring stable and uniform acoustic propagation characteristics of the ultrasonic waves along the coupling path. The focusing ultrasonic probe is positioned above the steel ball. A spherical piezoelectric crystal or acoustic lens structure at the bottom of the probe converges the emitted sound beam to form a focal zone. This focal zone is aligned with the geometric center of the steel ball, ensuring sufficient sound field coverage intensity for the ultrasonic energy throughout the entire depth range inside the ball. The sound beam penetrates the steel ball along its axial direction, propagating from the incident surface to the bottom of the ball. At each scanning azimuth, the probe emits a broadband pulse excitation signal and receives the full-time ultrasonic reflection signal, simultaneously recording the incident spatial angle coordinates corresponding to that azimuth. This ultimately forms a full-depth ultrasonic reflection signal dataset covering the entire spherical surface of the bearing steel ball.

[0081] The planning and execution of the ultrasonic scanning path for bearing steel balls relies on a precision motion control platform. This platform controls a focused ultrasonic probe to perform multi-directional point-to-point rotational scanning along the outer surface of the steel ball according to a preset spatial angular step distance. The scanning motion is achieved by using the geometric center of the steel ball as the rotation reference point. The focused ultrasonic probe is controlled to perform discrete steps along both the longitude and latitude directions according to set angular step distances, while ensuring that the probe's acoustic beam axis always passes through the geometric center of the steel ball. The size of the spatial angular step distance is determined based on the geometric relationship between the steel ball diameter and the target defect size. The selection of the angular step distance should ensure a moderate overlap between the acoustic beam coverage areas of adjacent scanning directions to avoid detection blind spots.

[0082] At each scanning azimuth, a focused ultrasonic probe emits a broadband pulse excitation signal and receives the returned ultrasonic reflection signal. The received reflection signal includes full-time reflection components such as the front surface echo reflected from the incident surface of the steel ball, the internal scattered echo reflected from internal defects or microstructure interfaces of the sphere, and the bottom reflection echo reflected from the bottom of the sphere. Throughout the scanning process, the incident spatial angle coordinates corresponding to each scanning azimuth are recorded synchronously. The incident spatial angle coordinates are represented by both longitude and latitude angle components and are used for subsequent spatial location calculations of micro-defects. The reflection signals received from all scanning azimuths are converged to form a full-depth ultrasonic reflection signal dataset covering the entire spherical surface of the bearing steel ball.

[0083] like Figure 3 The diagram shows a full-depth ultrasonic reflection signal and its digital representation. The time-domain waveform of the full-depth ultrasonic reflection signal sequentially presents multiple characteristic signal components along the time axis. The time reference zero point corresponds to the triggering moment of the ultrasonic probe's emitted pulse. The first strong-amplitude echo is the front surface echo, formed by interface reflection when ultrasonic waves propagate from the coupling medium to the incident surface of the steel sphere. Following the front surface echo, signal components from inside the sphere are distributed within the time interval. The weaker echoes correspond to scattering signals from internal micro-defects, formed by the scattering effect when ultrasonic waves encounter micro-defect interfaces during propagation inside the sphere. Continuing along the time axis, a stronger-amplitude bottom-reflection echo appears, formed by the reflection of ultrasonic waves at the opposite spherical boundary after penetrating the full depth of the steel sphere. Following the bottom-reflection echo, multiple reverberant echoes from the spherical boundaries are sequentially arranged. These reverberant echoes are generated by the reciprocating reflection of ultrasonic waves between the inner walls of the sphere. The amplitude of each order of reverberant echo decreases progressively with the number of reflections. Figure 3The process of high-speed analog-to-digital converter discretely sampling continuous analog signals is also indicated. Each sampling point is distributed at equal intervals on the time axis. Each sampling point is assigned an absolute time coordinate relative to the trigger zero point, forming a digital time-domain amplitude sequence with complete time axis marking. The superposition components of internal defect echo and multiple spherical boundary reverberation echo in this sequence are arranged in flight time order determined by their respective acoustic propagation paths.

[0084] After acquiring the full-depth ultrasonic reflection signal of the steel ball, high-speed analog-to-digital conversion sampling is required to complete signal digitization and time axis marking, discretizing the continuous analog reflection signal into a digital amplitude sequence with equal intervals. The quantization bit depth of the analog-to-digital converter is selected according to the dynamic range requirements of the tested signal. The quantization bit depth must ensure that the amplitude information of the weak defect echo is not submerged by the quantization noise floor under the condition that strong echoes from the front surface and weak echoes from internal defects coexist. After digitization, the trigger time of the ultrasonic probe's emitted pulse is used as the zero point of the time reference, and an absolute time coordinate is assigned to each sampling point in the digital amplitude sequence according to the sampling interval. The assignment method of the absolute time coordinate is as follows: the time coordinate corresponding to the first sampling point is equal to the time after the trigger time by one sampling interval, and the time coordinates of subsequent sampling points are increased by one sampling interval in turn. Through the above time axis marking process, a time-domain amplitude sequence with complete time axis marking is generated. The time-domain amplitude sequence simultaneously contains the superposition components of internal defect echoes and multiple reverberation echoes from the spherical boundary. These superposition components are arranged on the time axis according to the flight time order determined by their respective acoustic propagation paths.

[0085] The signal acquisition and digitization process in this step obtains a time-domain amplitude sequence with the trigger time as the time reference for each scanning direction. The time-domain amplitude sequences of all directions, together with the corresponding incident spatial angle coordinates, constitute the original data basis for subsequent signal processing and defect location analysis. The liquid immersion coupling method eliminates the problem of inconsistent contact coupling between the probe and the curved surface of the steel ball. The configuration of focusing the sound beam aligned with the geometric center of the steel ball makes the sound field intensity distribution throughout the entire depth of the sphere tend to be uniform. The multi-directional point-by-point rotation scanning strategy ensures that all areas inside the sphere are effectively detected and covered. High-speed digital sampling and precise time axis marking provide data quality assurance for subsequent fine processing of time-domain signals.

[0086] Step 2: Based on the geometric diameter parameters and internal sound velocity of the steel ball, construct the path tracing logic for multiple reflections at the spherical boundary, calculate the transit time of the ultrasonic wave propagating back and forth within the spherical boundary, and generate the temporal distribution matrix of the theoretical reverberation nodes.

[0087] After acquiring and digitizing the full-depth ultrasonic reflection signal of the bearing steel ball, it is necessary to establish a theoretical model for the multiple reflection propagation of ultrasonic waves under spherical boundary conditions in order to accurately predict the location of the reverberant echo at the spherical boundary in the time-domain amplitude sequence. Step 2 revolves around the construction of the logic for tracing the multiple reflection paths at the spherical boundary, and sequentially completes the acquisition of acoustic parameters, the establishment of the spherical reflection geometric model, the calculation of the sound path of multiple reflection paths, the transit time conversion, and the generation of the theoretical reverberant node time-series distribution matrix.

[0088] Specifically, the first step is to obtain the relevant acoustic parameters, namely the nominal geometric diameter of the bearing steel ball and the longitudinal wave velocity of the steel ball material. The nominal geometric diameter is obtained from the product specification document of the bearing steel ball under test or through actual measurement using a high-precision outer diameter measuring instrument. This parameter is used for calculating the sound path length in the subsequent spherical reflection geometric model. The longitudinal wave velocity of the steel ball material is retrieved from a material acoustic property database based on the steel ball's material grade, or measured through a longitudinal wave velocity calibration experiment on a standard test block. The longitudinal wave velocity reflects the speed of the ultrasonic longitudinal wave along the propagation direction within the steel ball material and is a key parameter for converting the sound path length into time of flight. Simultaneously, the velocity of the liquid-immersed coupling medium must also be obtained. This velocity is obtained by consulting a liquid medium acoustic property handbook for the corresponding velocity value at the current testing environment temperature, or by measuring the transit time of the ultrasonic wave in the coupling medium on a standard flat plate test block of known thickness and dividing the thickness by the transit time for calibration. For example, when the immersion coupling medium is deionized water that has undergone degassing treatment and the ambient temperature is 25 degrees Celsius, the longitudinal wave velocity of water, according to acoustic characteristic data, is approximately 1497 meters per second; when the ambient temperature is 20 degrees Celsius, the longitudinal wave velocity of water is approximately 1480 meters per second. The acoustic velocity parameter of the immersion coupling medium is used to subsequently calculate the time component of ultrasonic wave propagation in the coupling medium.

[0089] Furthermore, a geometric model for ultrasonic wave propagation under spherical boundary conditions is established. A three-dimensional coordinate system is established with the geometric center of the bearing steel ball as the origin, and the nominal geometric diameter parameter of the steel ball is used as the coordinate system. Half of the sphere's radius ,Right now The mathematical equation defining the boundary of a sphere is as follows: ,in , , Let be the coordinate components in a three-dimensional spatial coordinate system with the geometric center of the steel sphere as the origin. This equation describes the geometric constraints satisfied by all reflection points on the spherical boundary of the inner wall of the steel sphere. In the established geometric model, the complete ray tracing path of the ultrasonic wave is defined as follows: after entering the sphere from the incident point, the ultrasonic wave is reflected successively through the inner wall of the sphere. Specifically, the ultrasonic beam originates from a focusing ultrasonic probe, propagates in the liquid-immersed coupling medium to the incident surface of the steel sphere, and then refracts into the interior of the sphere. After entering the sphere, the sound wave propagates along the refraction direction until it reaches the opposite spherical boundary of the inner wall of the sphere. After a single reflection at the opposite spherical boundary, the propagation direction changes and continues to propagate inside the sphere. This repeated reflection between the inner walls of the sphere forms multiple reverberant propagation paths.

[0090] like Figure 4 The diagram illustrates the path tracing of multiple reflections at a spherical boundary. This path tracing demonstrates the propagation and reflection of ultrasonic waves within the sphere from the perspective of the cross-section of the bearing steel ball. The center of the sphere's cross-section is marked as the center of the sphere. A focusing ultrasonic probe is positioned above the sphere, and the sound beam travels through the coupling medium to the incident point on the sphere before entering the interior of the sphere. Figure 4 The diagram illustrates two typical reflection trajectories: the central sound ray path and the edge sound ray path. The central sound ray propagates along the beam axis. Since the axis passes through the center of the sphere, the angle of incidence of the central sound ray at the inner wall of the sphere is zero degrees. The reflection direction returns along the original path, forming a reciprocating propagation path along the diameter. The path length between two adjacent reflections is equal to the nominal geometric diameter of the steel sphere. The edge sound ray deviates from the beam axis by a deflection angle. The angle of incidence at the inner wall of the sphere is not zero, and the reflection direction deviates from the incident direction. This causes the sound ray to propagate along a polygonal broken-line path between the inner walls of the sphere. The path length between two adjacent reflections is the chord length of the sphere, the value of which is determined by both the deflection angle and the curvature of the sphere. Figure 4 The positions of each reflection node on the sphere are marked, and the lines connecting the reflection nodes form a complete ray tracing path sequence. By establishing reflection path tracing for the center sound ray and the edge sound ray respectively, the cumulative sound path length of the ultrasonic wave after multiple reflections at the spherical boundary inside the sphere can be calculated, providing an accurate geometric basis for subsequent transit time conversion and theoretical reverberation time node prediction.

[0091] Because the beam axis of a focused ultrasound probe always passes through the geometric center of the steel sphere during scanning, the angle between the incident direction of the sound wave at the spherical boundary and the normal direction of the sphere is determined by both the curvature of the sphere and the direction of sound beam propagation. When the beam axis passes through the center of the sphere, the incident angle of the central sound ray propagating along the axis at the inner wall of the sphere is zero, and the reflection direction returns along the original incident direction, forming a reciprocating propagation path in the diameter direction. However, the actual focused sound beam has a certain angle of attack, and the incident angle of the edge sound rays deviating from the axis within the beam cross-section at the spherical boundary is not zero. The reflection direction deviates from the incident direction, causing the reflection path of the edge sound rays between the inner walls of the sphere to no longer reciprocate along the diameter direction, but instead forming a reflection path distributed along a polygonal zigzag line along the inner wall of the sphere. In the construction of the ray tracing path, reflection path tracing sequences are established for the central sound ray and typical edge sound rays respectively. For the central sound ray, the path length between each adjacent reflection is equal to the nominal geometric diameter of the steel sphere. For edge sound ray, based on the curvature of the sphere and the angle of the sound ray's deviation from the axis, the spatial geometric relationship between the incident angle and the reflection angle of each reflection is determined using the geometric relationship of the sphere. Then, the chord length between each two adjacent reflections of the edge sound ray at the spherical boundary is calculated as the sound path length.

[0092] When calculating the sound path of multiple reflections, it is necessary to determine the spatial geometric relationship between the incident angle and the reflection angle of each reflection based on the curvature of the sphere, calculate the sound path length between every two adjacent reflections at the spherical boundary within the sphere, and construct a path propagation sequence covering multiple reflection orders. For the center sound ray, let the nominal geometric diameter of the bearing steel ball be... The path length between each two adjacent reflections of the central sound ray from the spherical boundary inside the sphere is . ,in This indicates the nominal geometric diameter of the steel balls in the bearing being inspected. (Center sound ray) The total acoustic path length inside the sphere corresponding to the first-order reflected echo is ,in This indicates the order number of the reflection from the spherical boundary, incrementing from one positive integer. Each order corresponds to one complete round-trip propagation of the ultrasonic wave between the front surface of the sphere and the rear wall of the opposite side. For an angle offset from the axis of θ... The edge of the voice, among which The deflection angle between the edge ray direction and the beam axis direction, and the chord length sound path between each two adjacent reflections from the spherical boundary. Obtained through spherical geometry calculations, specifically the following calculation method: ,in The radius of the steel ball is , The deflection angle between the edge sound ray direction and the sound beam axis direction is given by a formula derived from the geometric relationship between the length of the inscribed chord of a sphere and the central angle. When the deflection angle... When the chord length is zero, it degenerates into diameter. The acoustic path is consistent with that of the center vocal tract. Indicates the deflection angle as The path length of the edge acoustic ray between two adjacent reflections inside the sphere is recorded. The path propagation sequence is indexed by the reflection order number, and the path length value corresponding to each order of reflection is recorded sequentially. The maximum order number of the sequence is determined according to the acoustic attenuation characteristics of the ultrasonic wave propagating in the spherical material. When the estimated amplitude of a certain order reflection echo attenuates to below the electrical noise floor level of the ultrasonic detection system, this order is taken as the truncated order of the path propagation sequence.

[0093] Once the sound path length is obtained, the transit time can be calculated. Based on the sound path lengths corresponding to each order of reflection in the constructed path propagation sequence, and combined with the longitudinal wave velocity parameters of the steel ball material, the sound path lengths of each order of reflection are converted into corresponding single transit time components. For the center sound ray... The cumulative sound path inside the sphere is the order of reflected echoes. The corresponding cumulative transit time inside the sphere The calculation method is to divide the cumulative sound path by the longitudinal wave velocity of the steel ball material. ,Right now ,in Indicates the center vocal tract. The cumulative transit time of the reflected echo from the boundary of a sphere of order propagating inside the sphere. This represents the longitudinal wave velocity of the steel ball material. However, the complete transit time of the reflected echo to the probe receiving surface must be superimposed with the two-way propagation time component of the ultrasonic wave in the liquid immersion coupling medium. Let the liquid immersion coupling path length of the ultrasonic wave from the probe emitting surface to the incident surface of the steel ball be... The velocity of sound in the liquid immersion coupling medium is ,in This indicates the thickness of the coupling medium layer between the probe's emitting surface and the incident surface of the steel ball. Let represent the propagation speed of the longitudinal wave in the liquid-immersed coupling medium. Then, the two-way propagation time of the ultrasonic wave in the coupling medium is: Starting from the moment the ultrasonic wave is incident on the front surface of the sphere, the cumulative sound path time corresponding to the time taken by the ultrasonic wave to be reflected from the interior of the sphere to the boundary of each order of the sphere and back to the receiving surface of the probe is successively added together to obtain the [number of steps]. The cumulative transit time of the echo reflected from the spherical boundary of the order reaches the receiving surface of the probe. The expression is ,in Indicates the center vocal tract. The cumulative transit time of the spherical boundary reflected echo from the moment the probe emits the pulse until the echo reaches the probe's receiving surface is defined as follows: For example, for a bearing steel ball with a nominal geometric diameter of 9.525 mm, a longitudinal wave velocity of 5900 m / s, a coupling medium layer thickness of 15 mm from the probe's emitting surface to the steel ball's incident surface, and a coupling medium velocity of 1480 m / s, the cumulative transit time of the first-order spherical boundary reflected echo is approximately 20.27 + 3.23 = 23.50 microseconds, the cumulative transit time of the second-order spherical boundary reflected echo is approximately 20.27 + 6.46 = 26.73 microseconds, and the transit times of subsequent orders increase progressively.

[0094] Finally, a theoretical reverberation node time-series distribution matrix is ​​generated. The cumulative transit time values ​​corresponding to the reflected echoes from the spherical boundary of each order are arranged in ascending order of reflection, forming a one-dimensional theoretical reverberation time node sequence. Each element in the one-dimensional theoretical reverberation time node sequence represents the theoretical occurrence time of a reverberant echo of a specific reflection order in the time-domain amplitude sequence. Based on the one-dimensional time node sequence, each theoretical reverberation time node is labeled with its corresponding reflection order number and the estimated echo amplitude attenuation weight. The reflection order number is directly taken from the order index of the path propagation sequence.

[0095] The estimation of echo amplitude attenuation weight comprehensively considers three factors: first, the proportion of energy reflection loss due to acoustic impedance mismatch when the ultrasonic wave reflects each time it passes through the spherical boundary; and second, the amplitude reflection coefficient of the steel ball-coupling medium interface. express, ,in The acoustic impedance of the steel ball material. The first factor is the acoustic impedance of the coupling medium; the second is the acoustic attenuation of the ultrasonic wave during propagation in the steel ball material due to scattering and absorption within the material, expressed as the material attenuation coefficient. This indicates that the unit is nanops per meter, and the corresponding sound path is... The attenuation factor is Thirdly, the geometric attenuation caused by sound beam diffusion is approximated by the reciprocal of the propagation distance. Amplitude attenuation weight of order reflected echo The calculation expression is as follows ,in For the first The cumulative path length of the first-order reflected echoes. The cumulative path length of the first-order reflected echo. Indicates that the ultrasound passes through The cumulative reflection amplitude loss at the spherical boundary interface. The echo amplitude attenuation weight uses the amplitude of the first-order reflected echo as a normalized reference. The normalized amplitude attenuation weight of subsequent reflected echoes is expressed as a ratio relative to the amplitude of the first-order reflected echo, i.e., the... Normalized decay weights of order ,in This is the first-order amplitude attenuation weight. After normalization, it is always equal to 1, and subsequent orders are always equal to 1. The value varies The reverberation time nodes are increased and monotonically decreased. A multidimensional temporal distribution matrix is ​​constructed by combining the reverberation time nodes, reflection order numbers, and amplitude attenuation weights. Each row in the temporal distribution matrix corresponds to a reverberation event of a specific reflection order and its temporal characteristic parameters. The column dimension of the temporal distribution matrix includes three fields: the first column is the cumulative transit time value, the second column is the reflection order number, and the third column is the amplitude attenuation weight. The number of rows in the temporal distribution matrix is ​​equal to the truncation order value of the path propagation sequence.

[0096] The process of tracing the multiple reflection paths at the spherical boundary and generating the theoretical reverberation node time-series distribution matrix in this step transforms the physical laws governing the multiple reverberation propagation of ultrasonic waves inside the bearing steel ball into a digital theoretical template that can be compared and matched with measured time-domain data. The path tracing logic is constructed based on precise spherical geometry, avoiding model errors caused by simplified approximations of the spherical boundary reflection paths. The stepwise superposition calculation method of multi-order transit times ensures the cumulative accuracy of the theoretical reverberation time nodes. The prediction of amplitude attenuation weights provides a priori reference for signal strength matching during subsequent reverberation component stripping. The entire time-series distribution matrix provides complete reference template data for dynamic time warping and alignment in step 3.

[0097] Step 3: Dynamically time-warp and align the temporal distribution matrix of the theoretical reverberation nodes with the temporal amplitude sequence, strip the boundary reverberation components from the temporal amplitude sequence, and extract candidate echo envelopes containing internal micro-defect features.

[0098] After obtaining the temporal distribution matrix of the theoretical reverberation nodes and the measured time-domain amplitude sequence, it is necessary to accurately align and match the theoretically predicted reverberation time distribution of the spherical boundary with the actual reverberation echoes appearing in the measured signal. Then, the spherical boundary reverberation components are completely stripped from the measured time-domain amplitude sequence, and candidate echo envelopes that may contain scattering features of internal micro-defects are extracted from the stripped residual signal. This step will sequentially complete three stages: dynamic time warping alignment, adaptive cancellation stripping of boundary reverberation components, and candidate echo envelope extraction.

[0099] Specifically, dynamic time warping and alignment are first performed, using each reverberation time node in the temporal distribution matrix of the theoretical reverberation nodes as a reference template sequence, and local extrema points in the measured time-domain amplitude sequence as the feature sequence to be matched. The reference template sequence is constructed by extracting the cumulative transit time values ​​in the first column of the temporal distribution matrix row by row, and the sequence length is equal to the number of rows in the temporal distribution matrix, i.e., the truncation order value.

[0100] The construction process of the feature sequence to be matched includes: detecting local extrema in the measured time-domain amplitude sequence, extracting the time coordinates of all local maxima points whose absolute amplitude exceeds a preset extrema detection threshold, and constructing the feature sequence to be matched. The preset extrema detection threshold is taken as the percentage value of the peak amplitude of the front surface echo in the measured time-domain amplitude sequence. The percentage value is selected based on the electrical noise floor level of the detection system, so that the extrema detection threshold is higher than the peak amplitude of the electrical noise floor. A nonlinear elastic time mapping path is established between the reference template sequence and the feature sequence to be matched using a dynamic time warping algorithm.

[0101] like Figure 5 The diagram shown illustrates boundary reverberation component stripping and candidate echo envelope extraction. Figure 5 The process of boundary reverberation component stripping and candidate echo envelope extraction is presented in three stages. The first stage is the time-domain waveform of the original measured signal, which includes front surface echoes, weak internal defect scattering echoes, and multiple spherical boundary reverberation echoes with successively decaying amplitudes. Each reverberation component occupies a significant portion of the signal's energy. The second stage is the residual time-domain signal after adaptive cancellation stripping. In this stage, the spherical boundary reverberation components have undergone sequential cancellation operations according to the time window determined by dynamic time warping alignment. The signal within the time interval where the reverberation echo is located is effectively suppressed to near the background noise level, while the signal components outside the reverberation time window remain intact. The internal micro-defect scattering echoes are clearly preserved in the stripped residual signal. The third stage involves extracting the candidate echo envelopes. A signal energy detection threshold based on the root mean square amplitude of the background noise is set for the residual signal. Valid echo segments exceeding the detection threshold are identified and marked. Subsequently, a Hilbert transform is applied to each valid echo segment to extract the amplitude envelope curve. The envelope curve exhibits a smooth contour shape that rises from the background level to the peak and then falls back. The peak position and start and end time coordinates of the envelope are recorded. After clustering and screening, the candidate echo envelopes of suspected internal micro-defect scattering sources are output to subsequent processing stages.

[0102] The core computational process of the Dynamic Time Warping algorithm includes: constructing a two-dimensional cumulative distance matrix with row indices for elements of the reference template sequence and column indices for elements of the feature sequence to be matched. Each element in the cumulative distance matrix represents the cumulative cost of matching the time node at the corresponding row index position in the reference template sequence with the extreme point at the corresponding column index position in the feature sequence to be matched. The specific calculation method for the cumulative cost is as follows: Let the cumulative distance matrix be... The row index is The column index is Reference template sequence number The elements are The feature sequence to be matched The elements are Then the local distance Cumulative costs ,in As initial conditions. This indicates that the minimum cumulative cost among three adjacent predecessor elements is selected, ensuring that each expansion step chooses the matching path with the minimum global cost. The local distance metric for elements in the two-dimensional cumulative distance matrix is ​​the absolute value of the difference between two time values. The recursive filling direction of the cumulative distance matrix is ​​to advance element by element from the lower left corner to the upper right corner. Each step is limited to three choices: stepping one unit along the row direction, stepping one unit along the column direction, or stepping one unit simultaneously along the diagonal direction. The stepping direction that minimizes the cumulative distance cost is selected from these three choices. After filling all elements of the cumulative distance matrix, the matrix is ​​backtracked from the upper right corner to the lower left corner, and the optimal alignment path that minimizes the cumulative distance cost is searched node by node. Each node on the optimal alignment path establishes a one-to-one correspondence between a theoretical reverberation time node in the reference template sequence and a measured extreme point in the feature sequence to be matched, completing the one-to-one time alignment between the theoretical reverberation time node and the corresponding boundary reverberation echo peak in the measured time-domain amplitude sequence. The nonlinear elastic time mapping path established by the dynamic time warping algorithm can tolerate the systematic time offset and non-uniform time stretching caused by factors such as the small deviation between the actual diameter and nominal diameter of the steel ball, the temperature drift of the material sound velocity, and the fluctuation of the coupling medium layer thickness between the theoretical prediction time and the measured signal time, thereby achieving robust alignment and matching between the theoretical reverberation template and the measured signal.

[0103] The removal of boundary reverberation components employs an adaptive cancellation method. Based on the optimal alignment path obtained through dynamic time warping alignment, the time window location and duration range of each order of spherical boundary reverberation echo are located in the measured time-domain amplitude sequence. For each corresponding node on the optimal alignment path, the time window containing the reverberation echo is extended forward and backward by half the duration of the reverberation echo, centered on the time coordinate of the measured extreme point in the feature sequence to be matched. The duration of the reverberation echo is estimated based on the pulse width of the focused ultrasound probe and the pulse broadening effect during the reflection process of the spherical boundary; the duration is generally set to 1.5 to 2 times the probe's pulse width. Within each located reverberation time window, a local reverberation signal estimation template is constructed based on the amplitude attenuation weights of the corresponding order in the time-series distribution matrix.

[0104] The construction method of the local reverberation signal estimation template includes: using the waveform of the first-order spherical boundary reflection echo in the measured time-domain amplitude sequence within its time window as the base template waveform, multiplying the base template waveform by the amplitude attenuation weight value corresponding to the target order in the time-series distribution matrix, and shifting it on the time axis to the center position of the target order reverberation time window to obtain the local reverberation signal estimation template of the target order. When constructing the local reverberation signal estimation template, the amplitude of the base template waveform also needs to be fine-tuned to adapt to the actual amplitude level of the corresponding reverberation echo in the measured signal. The fine-tuning coefficient is obtained by fitting the amplitude ratio between the measured signal and the estimation template within the target order reverberation time window using the least squares method. The specific calculation method is as follows: assuming the sampling value of the measured time-domain amplitude sequence within the target order reverberation time window is... The corresponding sampled value of the local reverberation signal estimation template is , The fine-tuning coefficient is the index of the sampling point within the time window. By minimizing the sum of squared residuals Please solve this problem. Taking the derivative and setting it to zero, we get... Adjust the coefficients Multiply by the local reverberation signal estimation template The final reverberation estimate signal after amplitude adaptation is obtained and used for cancellation operation. The signal components of the measured time-domain amplitude sequence within each reverberation time window are adaptively cancelled with the corresponding local reverberation signal estimation template.

[0105] The adaptive cancellation operation includes: within each reverberation time window, subtracting the corresponding amplitude value of the local reverberation signal estimation template from the amplitude value of the measured time-domain amplitude sequence at each sampling point. The result of the subtraction operation replaces the original amplitude value of the measured time-domain amplitude sequence within the reverberation time window. The cancellation operation is performed sequentially according to the reflection order from low to high; that is, the first-order spherical boundary reverberation component is cancelled first, then the second-order spherical boundary reverberation component is cancelled, and so on, until all reverberation components corresponding to the truncation order of the path propagation sequence are cancelled. This sequential arrangement of cancellation orders ensures that the time windows of higher-order reverberation echoes are located and cancelled based on the signal after the lower-order reverberation components have been stripped, avoiding the impact of temporal overlap interference between adjacent-order reverberation echoes on the cancellation accuracy.

[0106] The process of extracting candidate echo envelopes involves setting a signal energy detection threshold in the residual time-domain signal after removing the boundary reverberation components, and identifying valid echo segments in the residual time-domain signal that exceed the signal energy detection threshold. The signal energy detection threshold is set based on the root mean square amplitude of the background noise in the residual time-domain signal after removing the reverberation components. The threshold is a preset multiple of the root mean square amplitude of the background noise in the residual time-domain signal. The preset multiple must ensure that the detection threshold is higher than the peak amplitude of the background noise but lower than the estimated amplitude of the echo scattered by the smallest target micro-defect. Instantaneous signal energy is detected by sliding through time windows in the residual time-domain signal. When the instantaneous signal energy within a certain time window exceeds the signal energy detection threshold, the signal segment covered by that time window is marked as a valid echo segment. The amplitude envelope curve of each identified valid echo segment is extracted using the Hilbert transform method. A Hilbert transform operation is performed on each valid echo segment to obtain an analytic signal, and the magnitude of the analytic signal is taken as the amplitude envelope curve. After extracting the amplitude envelope curve, isolated non-defect random interference is eliminated through envelope peak clustering analysis.

[0107] Envelope peak clustering analysis includes: detecting the peak positions of the amplitude envelope curves of each effective echo segment; grouping envelope peaks with a time interval less than a preset clustering interval threshold into the same cluster (the preset clustering interval threshold can be three times the pulse width emitted by the focused ultrasound probe); and identifying isolated peaks that exist alone and whose intervals with any other envelope peak are greater than the preset clustering interval threshold as spurious signals generated by non-defect random interference and removing them. The remaining effective echo segments after clustering are then selected as echo signals suspected to be internal micro-defect scattering sources. The selected effective echo segments, along with their time coordinate information, are extracted to form candidate echo envelopes containing internal micro-defect characteristics. Each effective echo segment in the candidate echo envelope carries its start and end time coordinates and peak time coordinates on the time axis of the original time-domain amplitude sequence.

[0108] This step, through dynamic time warping alignment, boundary reverberation component stripping, and candidate echo envelope extraction, effectively separates the spherical boundary reverberation components, which occupy significant energy in the measured time-domain amplitude sequence, from the potentially existing internal micro-defect scattered echoes. The nonlinear elastic matching characteristics of the dynamic time warping algorithm ensure that the theoretical reverberation template can still accurately locate the reverberation echo position when faced with fluctuations in actual detection conditions. The adaptive cancellation operation preserves the signal integrity in the non-reverberation time interval while successively stripping the reverberation components. The joint screening strategy of signal energy detection threshold and envelope peak clustering analysis effectively identifies candidate defect echoes while suppressing the misjudgment rate of random interference signals, providing a high-quality input signal for the subsequent refinement of the spiking neural network model.

[0109] Step 4: Input the candidate echo envelope into the pulse neural network model that integrates the adaptive threshold membrane potential mechanism to encode the temporal pulse characteristics of the candidate echo envelope and suppress residual high-frequency electrical noise to obtain a pure internal micro-defect echo pulse sequence.

[0110] After the dynamic time warping and boundary reverberation component removal in step 3, the extracted candidate echo envelope may still contain high-frequency electrical noise components inherent to the ultrasonic testing system. The spectra of these high-frequency electrical noise components partially overlap with the spectra of the internal micro-defect scattering echo signals, making complete separation difficult using traditional frequency domain bandpass filtering. This step employs a spiking neural network model incorporating an adaptive threshold membrane potential mechanism to perform time-series pulse coding and high-frequency electrical noise suppression on the candidate echo envelope. Utilizing the time-domain integral filtering characteristics and adaptive threshold adjustment capability of the spiking neural network model, a pure internal micro-defect echo pulse sequence is extracted from the candidate echo envelope.

[0111] From an architectural design perspective, the spiking neural network model comprises a multi-layered network architecture consisting of an input coding layer, a pulse processing hidden layer, and an output decoding layer. The input coding layer receives candidate echo envelope signals and converts continuous amplitude signals into discrete pulse sequences. The pulse processing hidden layer performs time-domain integration and feature extraction on the input pulse sequence. The output decoding layer converts the output pulses from the pulse processing hidden layer into temporal pulse features that can be used for subsequent defect localization. The number of neurons in the input coding layer is equal to the number of sampling points in the candidate echo envelope signal within a single processing time window; each neuron in the input coding layer corresponds to a discrete time step in the candidate echo envelope signal. The number of layers in the pulse processing hidden layer and the number of neurons in each layer are configured according to the temporal complexity of the candidate echo envelope signal. The number of layers in the pulse processing hidden layer is generally two to three, and the number of neurons in each layer is an integer value between 0.5 and 2 times the number of neurons in the input coding layer. The number of neurons in the output decoding layer is the same as the number of neurons in the last layer of the pulse processing hidden layer. Each neuron in the output decoding layer outputs the received processed pulse signal as a temporal pulse sequence.

[0112] The input coding layer is responsible for signal conversion, transforming the amplitude-time sequence signal of the candidate echo envelope into a pulse sequence that is input to the pulse processing hidden layer. The input coding layer uses rate coding for amplitude-to-pulse conversion: at each discrete time step, the corresponding neuron in the input coding layer normalizes the amplitude value of the candidate echo envelope at the current time step to a range between zero and one, with the global peak amplitude of the candidate echo envelope serving as the reference for normalization. The normalized amplitude value serves as the probability value for the input coding layer neuron to generate an output pulse at the current time step; a higher amplitude value indicates a greater probability of pulse generation, and a lower amplitude value indicates a lower probability of pulse generation. When the amplitude of the candidate echo envelope is at a high level at a certain time step, the input coding layer generates high-density pulse firing at that time step, reflecting the strong signal characteristics of the internal micro-defect scattering echo; when the amplitude of the candidate echo envelope is at a low level at a certain time step, the input coding layer generates low-density or no pulse firing at that time step, corresponding to signal intermittent intervals or weak noise intervals. Using the rate coding method described above, the continuous amplitude information of the candidate echo envelope is mapped to the firing density distribution of the time-domain pulse sequence.

[0113] The dynamic mechanism of each spiking neuron in the hidden layer of the spiking processing layer is constructed using a leakage integral firing model. Each spiking neuron maintains a membrane potential state variable that evolves dynamically over time. The update of the membrane potential state variable at each discrete time step follows the following dynamic process: At the current time step, the membrane potential state variable first undergoes leakage decay based on the membrane potential value of the previous time step. The degree of leakage decay is controlled by the membrane potential leakage coefficient. Take a constant value between zero and one. This represents the scaling factor that is retained when the membrane potential of a spiking neuron is transmitted from the previous time step to the current time step. The closer the value is to a given membrane potential, the longer the time memory. The closer the value of is to zero, the faster the membrane potential leakage rate. For example, in an ultrasonic testing signal processing scenario where the micro-defect echo pulse width is ten sampling intervals, the membrane potential leakage coefficient... A value of 0.85 can be chosen to ensure that the effective memory time of the membrane potential covers approximately 6 to 7 sampling intervals, thus enabling effective time-domain integration of the continuous amplitude information of micro-defect echo pulses. After leakage attenuation, the membrane potential state variable is superimposed with the synaptic weighted input current from the previous layer at the current time step. The synaptic weighted input current is equal to the weighted sum of the pulse values ​​generated at the current time step by all neurons in the previous layer that send pulses to the current spiking neuron, and the corresponding synaptic connection weights. The synaptic connection weights are pre-learned through the offline training phase of the spiking neural network model, and the training dataset consists of bearing steel ball detection signal samples containing known defect locations. After superimposing the synaptic weighted input current, the membrane potential state variable is compared with the firing threshold. When the accumulated membrane potential state variable exceeds the firing threshold, the spiking neuron generates an output pulse and resets the membrane potential to the baseline level, which is generally set to zero. When the membrane potential state variable does not reach the firing threshold, the spiking neuron does not generate an output pulse at the current time step, and the current value of the membrane potential state variable is retained to the next time step to continue participating in the leakage integration process.

[0114] An adaptive threshold membrane potential mechanism is introduced to set a dynamically adjustable firing threshold for each spiking neuron in the spiking processing hidden layer. This firing threshold is adaptively updated based on the historical firing frequency of the spiking neuron within a recent time window. This adaptive update mechanism remains in effect throughout the inference process of the spiking neural network model. Specifically, it includes defining a statistical time window and a pair of frequency boundary parameters for each spiking neuron. The length of the statistical time window is determined based on the typical duration of micro-defect echoes in the candidate echo envelope, and is typically 2 to 3 times the typical duration of the micro-defect echo. The frequency boundary parameters include an upper and a lower frequency limit. The upper frequency limit reflects the maximum reasonable firing frequency of the spiking neuron when processing micro-defect echo signals normally, while the lower frequency limit reflects the minimum expected firing frequency of the spiking neuron under effective signal input conditions. The specific values ​​of the upper and lower frequency limits are determined based on the statistical results of the firing frequencies of known defect samples during the training phase of the spiking neural network model.

[0115] During the operation of the spiking neural network model, the cumulative firing count of each spiking neuron within the most recent statistical time window is continuously counted. When the cumulative firing count of a spiking neuron within a preset statistical time window exceeds the upper frequency limit, it indicates that the spiking neuron is over-responding to high-frequency electrical noise components. The firing threshold of this spiking neuron is adjusted upward to raise its triggering condition, requiring a larger membrane potential accumulation in subsequent time steps to generate an output pulse, thereby suppressing the over-response to high-frequency electrical noise. The upward adjustment of the firing threshold is the current firing threshold multiplied by a preset upward adjustment coefficient, which is a constant value greater than one. The specific value is determined based on the parameter tuning experiments during the training phase of the spiking neural network model. When the cumulative firing count of a spiking neuron within a preset statistical time window is lower than the lower frequency limit, it indicates that the firing threshold of the spiking neuron is too high, resulting in insufficient response sensitivity to effective micro-defect echo signals. The firing threshold of this spiking neuron is adjusted downward to lower its triggering condition, allowing the spiking neuron to generate an output pulse with a lower membrane potential accumulation in subsequent time steps, thereby restoring the response sensitivity to effective signals. The downward adjustment of the firing threshold is calculated by multiplying the current firing threshold by a preset downward adjustment coefficient, which is a constant value between zero and one. Through this adaptive threshold adjustment mechanism, the firing threshold of each spiking neuron in the hidden layer of the pulse processing is dynamically and adaptively adjusted according to the statistical characteristics of the actual input signal during the operation of the spiking neural network model. During periods of high-frequency electrical noise activity, the trigger condition is automatically raised to suppress noise pulse output, while during periods of effective signal activity, the trigger condition is automatically lowered to preserve signal pulse output.

[0116] Temporal pulse coding and high-frequency electrical noise suppression work synergistically to achieve denoising. Specifically, the amplitude variation of the candidate echo envelope is converted into pulse firing time encoding through the input coding layer. The membrane potential integration characteristics of each spiking neuron in the pulse processing hidden layer are used to perform temporal aggregation and filtering of the input pulse sequence. The micro-defect scattering echo signal in the candidate echo envelope exhibits a continuous envelope shape with amplitude gradually increasing from the background noise level to the peak and then gradually decreasing. The input coding layer encodes this continuous envelope shape as a high-density pulse cluster concentrated within the echo duration. The membrane potential leakage integration mechanism of the spiking neurons in the pulse processing hidden layer performs temporal integration on the high-density pulse cluster. The continuous high-density input pulses from the micro-defect echo produce a cumulative effect on the membrane potential, causing the membrane potential to rise steadily and reach the firing threshold, thereby generating an output pulse corresponding to the temporal characteristics of the micro-defect echo. Unlike the temporal continuity of micro-defect scattering echoes, residual high-frequency electrical noise manifests as irregularly distributed, transient amplitude spikes in the candidate echo envelope. The input coding layer encodes these transient amplitude spikes into irregular, short-interval, scattered pulses. Based on the membrane potential leakage mechanism, the synaptic input currents corresponding to these irregular short-interval pulses, upon reaching the spiking neuron, rapidly decline due to the lack of continuous pulse input replenishment in subsequent time steps, failing to accumulate to the firing threshold and thus not triggering an output pulse. Even if the instantaneous amplitude of high-frequency electrical noise is high at certain moments, resulting in localized dense pulse inputs within a short period, the adaptive threshold membrane potential mechanism will detect the abnormal increase in the firing frequency of the spiking neuron and automatically raise the firing threshold, preventing noise-driven membrane potential accumulation from exceeding the new firing threshold. The effective temporal pulse sequence output from the pulse-processed hidden layer is determined as the encoded temporal pulse feature, which is received by the output decoding layer and passed to subsequent steps.

[0117] This step, through the synergistic effect of the multi-layer processing architecture of the spiking neural network model and the adaptive threshold membrane potential mechanism, effectively encodes and preserves the temporal pulse characteristics of the echoes scattered by internal micro-defects within the candidate echo envelope, while residual high-frequency electrical noise components are fully filtered out by the dual suppression mechanism of membrane potential leakage integral and adaptive threshold. The spiking neural network model, based on pulse timing encoding, is naturally suitable for the fine extraction of temporal features from ultrasonic echo signals. The membrane potential dynamics of the leakage integral model have good matching with the temporal continuous envelope shape of the ultrasonic echo signal. The adaptive threshold adjustment mechanism enables the spiking neural network model to maintain stable denoising performance under different signal-to-noise ratio conditions. The overall processing flow provides high-purity pulse feature data for the accurate determination of the absolute acoustic time of flight of micro-defects in subsequent steps.

[0118] Step 5: Decode the timing pulse characteristics to determine the absolute acoustic time of flight of the internal micro-defects, calculate the precise spatial coordinates of the micro-defects by combining the incident spatial angle of the ultrasonic probe, and output the ultrasonic detection results of the internal micro-defects of the bearing steel ball.

[0119] After processing by the spiking neural network model in step 4, a clean time-series pulse sequence encoding the time-domain characteristics of the internal micro-defect echoes is obtained. This step decodes and analyzes the time-series pulse characteristics output by the spiking neural network model to determine the absolute acoustic time of flight of each micro-defect echo. Combined with the incident spatial angle parameters of the ultrasonic probe at each scanning orientation, the radial depth distance of the micro-defect is converted into three-dimensional spatial coordinates. Finally, the detection information from each scanning orientation is summarized to generate and output the ultrasonic detection results of the internal micro-defects of the bearing steel ball.

[0120] Specifically, the process begins with temporal pulse feature decoding. This involves analyzing the pulse timing of the temporal pulse features output by the spiking neural network model to identify pulse clusters corresponding to each micro-defect echo and extract the starting emission time of each pulse cluster. The specific operation of pulse timing analysis is as follows: In the temporal pulse sequence output by the decoding layer of the spiking neural network model, the emission time coordinates of all output pulses are detected, and all pulses in the temporal pulse sequence are arranged in ascending order of emission time. Pulse cluster identification is then performed in the arranged pulse sequence: when the emission time interval between two adjacent pulses is less than a preset pulse cluster spacing threshold, these two pulses are grouped into the same pulse cluster; when the emission time interval between two adjacent pulses is greater than or equal to the preset pulse cluster spacing threshold, the latter pulse is taken as the starting pulse of a new pulse cluster. The preset pulse cluster spacing threshold is determined based on the typical duration of the micro-defect scattering echo in the candidate echo envelope, and is taken as the number of sampling point intervals corresponding to the typical duration of the micro-defect scattering echo. Through the above pulse cluster identification process, the temporal pulse sequence is divided into several discrete pulse clusters, each pulse cluster corresponding to a scattering echo event of an internal micro-defect. For each identified pulse cluster, its initial emission time is extracted. The initial emission time is the coordinate value of the emission time of the first pulse in the pulse cluster. The initial emission time reflects the time node when the micro-defect scattered echo signal first arrives at the input of the pulse neural network model and first triggers the output pulse after encoding and processing. It has a corresponding relationship with the arrival time of the leading edge of the micro-defect scattered echo.

[0121] Determining the absolute acoustic time of flight requires mapping the initial emission time of each pulse cluster back to the time axis coordinate system of the original time-domain amplitude sequence. Since the input to the spiking neural network model is the candidate echo envelope, which inherits the time axis coordinate markers of the original time-domain amplitude sequence, the coordinate values ​​of the pulse cluster's initial emission time on the processing time axis within the spiking neural network model can be traced back to the absolute time coordinates of the original time-domain amplitude sequence through the time mapping relationship of the input coding layer. After mapping back to the original time axis, the complete time interval from the moment the ultrasonic probe emits a pulse to the arrival of the scattered echo from the micro-defect at the probe's receiving surface is calculated. The complete time interval is equal to the absolute time coordinate value corresponding to the mapped pulse cluster's initial emission time minus the trigger time coordinate value of the ultrasonic probe's emitted pulse. Since the trigger time of the emitted pulse was set as the time reference zero point in step 1, the complete time interval is numerically directly equal to the mapped absolute time coordinate value. This complete time interval is defined as the absolute acoustic time of flight from the probe through the coupling medium into the steel ball to the micro-defect location and back. Absolute acoustic time of flight comprises two components: the two-way propagation time of ultrasound in the liquid-immersed coupling medium and the two-way propagation time of ultrasound inside the steel ball from the incident surface to the micro-defect location.

[0122] The first step in calculating the radial depth distance of micro-defects is to extract the arrival time of the front surface echo from the full-depth ultrasonic reflection signal to determine the one-way propagation time of the ultrasonic wave in the coupling medium. The front surface echo is the reflected echo generated on the surface of the steel ball when the ultrasonic wave propagates from the probe's emitting surface through the coupling medium to the incident surface of the steel ball. The arrival time of the front surface echo corresponds to the leading edge time coordinate of the first strong amplitude echo in the original time-domain amplitude sequence. The one-way propagation time of the ultrasonic wave in the coupling medium can be obtained from the arrival time of the front surface echo. ,in This represents the one-way propagation time of the ultrasonic wave from the emitting surface of the probe through the liquid-immersed coupling medium to the incident surface of the steel ball. Numerically, it is equal to half the coordinate value of the arrival time of the echo from the front surface. The two-way propagation time of the ultrasonic wave in the coupling medium is equal to... Based on the determined absolute acoustic time of flight of the micro-defect scattered echo. ,in This represents the total time it takes for the ultrasonic wave to travel from the moment the probe emits a pulse, through the coupling medium, into the interior of the steel ball, to the location of the micro-defect, and then be scattered back to the probe's receiving surface, minus the two-way propagation time of the ultrasonic wave in the coupling medium. The two-way propagation time of ultrasonic waves inside the steel ball from the incident surface to the micro-defect location was obtained. The calculation relationship is ,in This represents the two-way propagation time of an ultrasonic wave inside a steel sphere, from the incident surface to the micro-defect location and then back to the incident surface after scattering from the micro-defect location. It is the two-way propagation time of the ultrasonic wave inside the steel sphere. Longitudinal wave velocity of combined steel ball material Converted to radial depth distance of the micro-defect relative to the incident surface The conversion relationship is as follows: ,in This indicates the radial depth of the micro-defect location along the ultrasonic beam axis from the incident surface of the steel ball. This represents the longitudinal wave velocity of the steel ball material. For example, if the absolute acoustic time of flight of the echo scattered by a micro-defect... The arrival time of the front surface echo is 22.50 microseconds, and the arrival time of the front surface echo is 20.27 microseconds, corresponding to a one-way propagation time. If the time is 10.135 microseconds, then the two-way propagation time of the ultrasonic wave inside the steel ball is... Microseconds, combined with the longitudinal wave velocity of 5900 meters per second in the steel ball material, and the radial depth distance of the micro-defect. It is approximately 6.58 millimeters.

[0123] Calculating the three-dimensional spatial coordinates of a micro-defect requires obtaining the incident spatial angle parameters of the ultrasonic probe at the current scanning orientation. Combined with the radial depth distance, the micro-defect is then located to its precise spatial coordinates within the three-dimensional coordinate system inside the steel sphere. The incident spatial angle parameters are obtained from the longitude angular components recorded synchronously during the scanning process in step 1. and latitude angular components Composition, in which This indicates the angular coordinates of the current scan position in the longitude direction. This represents the angular coordinates of the current scan orientation in the latitude direction. In a three-dimensional Cartesian coordinate system with the geometric center of the steel ball as the origin, the precise spatial coordinates of the micro-defect are determined by the radial depth distance. Longitude angle and latitude angle This was determined jointly. Since the beam axis of the focused ultrasonic probe in step 1 always passes through the geometric center of the steel ball, the radial depth distance of the micro-defect... This corresponds to the depth along the sound beam axis from the incident surface of the steel ball towards the center of the ball. The coordinate components of the micro-defect in the three-dimensional Cartesian coordinate system are obtained through a transformation from spherical coordinates to Cartesian coordinates: subtract the radial depth distance from the radius of the steel ball. The radial vector length of the micro-defect from the center of the sphere is obtained. ,Right now ,in This represents the distance from the location of the micro-defect to the geometric center of the steel ball. This represents the nominal geometric diameter of the steel ball. The lateral coordinate components of the micro-defect in a three-dimensional Cartesian coordinate system. Vertical coordinate components and vertical coordinate components They are respectively , , ,in This represents the coordinate value of the micro-defect along the horizontal axis of the three-dimensional coordinate system. This represents the coordinate value of the micro-defect along the vertical axis of the three-dimensional coordinate system. This represents the coordinates of a micro-defect along the vertical axis of a three-dimensional coordinate system. When the same micro-defect is detected in multiple adjacent scanning orientations, the weighted average of the spatial coordinates calculated in each scanning orientation is taken as the final spatial coordinates of the micro-defect. The weighting coefficient is the normalized ratio of the number of pulses in the echo pulse cluster of the micro-defect in each scanning orientation. The more pulses, the higher the signal quality, and the greater the corresponding weight.

[0124] Finally, the test results are summarized and output. The spatial coordinates of micro-defects detected from each scanning orientation are integrated with the echo characteristic information to generate and output the ultrasonic test results of micro-defects inside the bearing steel ball. The test results include the following information: the precise spatial coordinates of each detected internal micro-defect in the three-dimensional coordinate system of the steel ball, the radial depth distance of the micro-defect, the absolute acoustic time of flight of the micro-defect scattered echo, the amplitude envelope peak value of the micro-defect scattered echo before and after processing by the pulse neural network model, the number of pulses in the pulse cluster corresponding to the micro-defect scattered echo, and the incident spatial angle coordinates of the scanning orientation of the detected micro-defect. The test results are arranged in descending order of the peak value of the micro-defect echo amplitude envelope, with micro-defects with larger echo amplitude envelope peak values ​​listed first, making it easier for inspectors to focus on defects with stronger reflected energy. If no micro-defects exceeding the judgment threshold are detected inside the inspected bearing steel ball, the test results output a defect-free judgment mark. The judgment threshold is set according to the quality grade standard of the bearing steel ball and the ultrasonic testing sensitivity calibration results.

[0125] This step, through a complete process of temporal pulse feature decoding, absolute acoustic time-of-flight determination, radial depth distance conversion, three-dimensional spatial coordinate calculation, and summary output of detection results, ultimately transforms the abstract temporal pulse features output by the spiking neural network model into spatially localized micro-defect information with clear physical meaning. The pulse cluster identification and initial emission time extraction methods ensure the accuracy of micro-defect echo arrival time determination. Precise calculation of the absolute acoustic time of flight and subtraction of the coupling medium propagation time eliminate the influence of detection system configuration differences on depth measurement results. The conversion from spherical coordinates to rectangular coordinates combined with a multi-directional weighted averaging strategy further improves the three-dimensional spatial localization accuracy of micro-defects. The structured output format of the detection results provides complete information support for subsequent bearing steel ball quality assessment and defect analysis.

[0126] Example 2

[0127] Reference Figure 6 This is the second embodiment of the present application, which provides a high-precision ultrasonic testing system for micro-defects inside bearing steel balls.

[0128] The system includes an ultrasonic signal acquisition module, a reverberation matrix generation module, an echo envelope extraction module, a pulse feature encoding module, and a defect location output module.

[0129] The ultrasonic signal acquisition module acquires the full-depth ultrasonic reflection signal of the bearing steel ball and records the corresponding incident spatial angle. It digitizes the full-depth ultrasonic reflection signal into a time-domain amplitude sequence containing internal defect echoes and multiple reverberant echoes from the spherical boundary, and marks the time axis coordinates.

[0130] The reverberation matrix generation module, based on the geometric diameter parameters and internal sound velocity of the steel ball, constructs a path tracing logic for multiple reflections at the spherical boundary, calculates the transit time of the ultrasonic wave propagating back and forth within the spherical boundary, and generates a temporal distribution matrix of theoretical reverberation nodes.

[0131] The echo envelope extraction module dynamically aligns the temporal distribution matrix of the theoretical reverberation nodes with the time-domain amplitude sequence, strips the boundary reverberation components from the time-domain amplitude sequence, and extracts candidate echo envelopes containing internal micro-defect features.

[0132] The pulse feature encoding module inputs the candidate echo envelope into a pulse neural network model that integrates an adaptive threshold membrane potential mechanism, encodes the temporal pulse features of the candidate echo envelope, and suppresses residual high-frequency electrical noise to obtain a pure internal micro-defect echo pulse sequence.

[0133] The defect location output module decodes the timing pulse characteristics to determine the absolute acoustic time of flight of the internal micro-defect, calculates the precise spatial coordinates of the micro-defect by combining the incident spatial angle of the ultrasonic probe, and outputs the ultrasonic detection results of the internal micro-defect of the bearing steel ball.

[0134] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the displayed or discussed mutual couplings or direct couplings or communication connections may be through some communication interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0135] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments under the guidance of this application without departing from the spirit and scope of protection of the claims. All of these variations are within the protection scope of this application.

Claims

1. A high-precision ultrasonic testing method for internal micro-defects in bearing steel balls, characterized in that, include: Acquire full-depth ultrasonic reflection signals of bearing steel balls and record the corresponding incident spatial angles. Digitize the full-depth ultrasonic reflection signals into a time-domain amplitude sequence containing internal defect echoes and multiple reverberant echoes from the spherical boundary, and mark the time axis coordinates. Based on the geometric diameter parameters and internal sound velocity of the steel ball, a path tracing logic for multiple reflections at the spherical boundary is constructed to calculate the transit time of the ultrasonic wave propagating back and forth within the spherical boundary and generate the temporal distribution matrix of the theoretical reverberation nodes. The temporal distribution matrix of the theoretical reverberation node is dynamically time-warped and aligned with the temporal amplitude sequence. The boundary reverberation component in the temporal amplitude sequence is removed, and the candidate echo envelope containing internal micro-defect features is extracted. The candidate echo envelope is input into a pulse neural network model that integrates an adaptive threshold membrane potential mechanism to encode the temporal pulse characteristics of the candidate echo envelope and suppress residual high-frequency electrical noise to obtain a pure internal micro-defect echo pulse sequence. Decode the timing pulse characteristics to determine the absolute acoustic time of flight of the internal micro-defects, combine the incident spatial angle of the ultrasonic probe to calculate the precise spatial coordinates of the micro-defects, and output the ultrasonic detection results of the internal micro-defects of the bearing steel ball.

2. The high-precision ultrasonic testing method for internal micro-defects in bearing steel balls according to claim 1, characterized in that, The acquisition of full-depth ultrasonic reflection signals from the bearing steel balls includes: A focused ultrasonic probe is used to apply broadband pulse excitation to the bearing steel ball in a liquid immersion coupling manner. The focal area of ​​the focused ultrasonic probe is aligned with the geometric center of the steel ball to cover the full depth of the sound field inside the ball. The focused ultrasound probe is controlled to perform multi-directional point-by-point rotation scanning along the outer surface of the steel ball according to a preset spatial angle step distance. At each scanning direction, the full-time reflection signal including the front surface echo, the internal scattered echo, and the bottom reflection echo of the sphere is received. The reflection signals from all directions are converged to form the full-depth ultrasound reflection signal.

3. The high-precision ultrasonic testing method for internal micro-defects in bearing steel balls according to claim 2, characterized in that, The process of digitizing the full-depth ultrasonic reflection signal into a time-domain amplitude sequence containing internal defect echoes and multiple spherical boundary reverberation echoes, and marking the time axis coordinates, includes: The full-depth ultrasonic reflection signal is sampled by high-speed analog-to-digital conversion, and the continuous analog reflection signal is discretized into a digital amplitude sequence with equal intervals. Using the trigger moment of the ultrasonic probe's emitted pulse as the zero point of the time reference, and assigning absolute time coordinates to each sampling point in the digital amplitude sequence according to the sampling interval, a time-domain amplitude sequence with complete time axis marking is generated. The time-domain amplitude sequence contains the superposition components of internal defect echoes and multiple spherical boundary reverberation echoes.

4. The high-precision ultrasonic testing method for internal micro-defects in bearing steel balls according to claim 1, characterized in that, The logic for constructing multiple reflection path tracing of the spherical boundary includes: Obtain the nominal geometric diameter parameters of the bearing steel balls and the longitudinal wave sound velocity parameters of the steel ball material; A geometric model for ultrasonic wave propagation under spherical boundary conditions is established. In the geometric model, a complete ray tracing path is defined for the ultrasonic wave after it enters the sphere from the incident point and is reflected successively by the inner wall of the sphere. The spatial geometric relationship between the incident angle and the reflection angle for each reflection is determined based on the curvature of the sphere. The acoustic path length between each two adjacent reflections of the ultrasonic wave within the sphere is calculated, and a path propagation sequence covering multiple reflection orders is constructed.

5. The high-precision ultrasonic testing method for internal micro-defects in bearing steel balls according to claim 4, characterized in that, The calculation of the transit time of the ultrasonic wave propagating back and forth within the spherical boundary includes: Based on the path lengths of each order of reflection in the constructed path propagation sequence, and combined with the longitudinal wave velocity parameters of the steel ball material, the path lengths of each order of reflection are converted into corresponding single transit time components. Starting from the incident ultrasonic wave on the front surface of the sphere, the cumulative sound path time corresponding to the reflection from the inside of the sphere to the boundary of each order of the sphere and back to the probe receiving surface is successively superimposed to obtain the cumulative transit time value of the echo reflected from the boundary of each order of the sphere to the probe receiving surface.

6. The high-precision ultrasonic testing method for internal micro-defects in bearing steel balls according to claim 5, characterized in that, The temporal distribution matrix of the generated theoretical reverberation nodes includes: The cumulative transit time values ​​corresponding to the reflected echoes from the spherical boundary of each order are arranged in order of reflection from low to high to form a one-dimensional theoretical reverberation time node sequence. For each theoretical reverberation time node, its corresponding reflection order number and estimated echo amplitude attenuation weight are labeled. The reverberation time node, reflection order number and amplitude attenuation weight are combined to construct a multi-dimensional time-series distribution matrix. Each row in the time-series distribution matrix corresponds to a reverberation event of a specific reflection order and its time-domain characteristic parameters.

7. The high-precision ultrasonic testing method for internal micro-defects in bearing steel balls according to claim 1, characterized in that, The step of dynamically time-warping and aligning the temporal distribution matrix of the theoretical reverberation nodes with the temporal amplitude sequence includes: Using each reverberation time node in the temporal distribution matrix of the theoretical reverberation nodes as a reference template sequence, and the local extreme points in the measured time-domain amplitude sequence as the feature sequence to be matched, a nonlinear elastic time mapping path is established between the reference template sequence and the feature sequence to be matched through a dynamic time warping algorithm. The optimal alignment path that minimizes the cumulative distance cost is searched node by node to complete the one-to-one time alignment between the theoretical reverberation time nodes and the corresponding boundary reverberation echo peaks in the measured time-domain amplitude sequence.

8. The high-precision ultrasonic testing method for internal micro-defects in bearing steel balls according to claim 7, characterized in that, The boundary reverberation components stripped from the time-domain amplitude sequence include: Based on the optimal alignment path obtained by dynamic time warping alignment, the time window position and duration range of each order of spherical boundary reverberant echo are located in the measured time domain amplitude sequence. Within each reverberation time window, a local reverberation signal estimation template is constructed based on the amplitude attenuation weights of the corresponding order in the time-series distribution matrix. The signal components of the measured time-domain amplitude sequence within each reverberation time window are adaptively canceled with the corresponding local reverberation signal estimation template, and the spherical boundary reverberation components are stripped off step by step.

9. The high-precision ultrasonic testing method for internal micro-defects in bearing steel balls according to claim 8, characterized in that, The extraction of candidate echo envelopes containing internal micro-defect features includes: In the residual time-domain signal after stripping the boundary reverberation component, a signal energy detection threshold is set to identify effective echo segments in the residual time-domain signal that exceed the signal energy detection threshold. The amplitude envelope curves of each identified effective echo segment are extracted. Isolated non-defect random interferences are eliminated by envelope peak clustering analysis, and effective echo segments suspected to be internal micro-defect scattering sources are screened out. The selected valid echo segments, along with their time coordinate information, are extracted to form a candidate echo envelope containing internal micro-defect features.

10. The high-precision ultrasonic testing method for internal micro-defects in bearing steel balls according to claim 1, characterized in that, The spiking neural network model includes a multi-layer network architecture consisting of an input encoding layer, a spiking processing hidden layer, and an output decoding layer. The input coding layer converts the amplitude timing signal of the candidate echo envelope into a pulse sequence and inputs it into the pulse processing hidden layer; Each spiking neuron in the hidden layer of the pulse processing adopts a leakage integral firing model. Each spiking neuron maintains a membrane potential state variable that evolves dynamically over time. When the accumulated membrane potential state variable exceeds the firing threshold, the spiking neuron generates an output pulse and resets the membrane potential to the baseline level.

11. The high-precision ultrasonic testing method for internal micro-defects in bearing steel balls according to claim 10, characterized in that, The adaptive threshold membrane potential mechanism includes: A dynamically adjustable firing threshold is set for each spiking neuron in the pulse processing hidden layer. The firing threshold is adaptively updated based on the historical pulse firing frequency of the spiking neuron in the recent time window. When the cumulative firing count of a spiking neuron within a preset statistical time window exceeds the upper frequency limit, the firing threshold of the spiking neuron is adjusted upward to raise its triggering condition; when the cumulative firing count is lower than the lower frequency limit, the firing threshold of the spiking neuron is adjusted downward to lower its triggering condition, thereby achieving dynamic adaptive adjustment of the firing threshold of each spiking neuron.

12. The high-precision ultrasonic testing method for internal micro-defects in bearing steel balls according to claim 11, characterized in that, Encoding the temporal pulse characteristics of candidate echo envelopes and suppressing residual high-frequency electrical noise includes: The amplitude variation of the candidate echo envelope is converted into pulse firing time encoding through the input coding layer, and the time-domain aggregation and filtering of the input pulse sequence is performed by utilizing the membrane potential integral characteristics of each spiking neuron in the pulse processing hidden layer. Based on the membrane potential leakage mechanism, the irregular short-interval pulses corresponding to the residual high-frequency electrical noise are integrally attenuated to prevent them from accumulating to the firing threshold; the effective time-series pulse sequence output by the pulse-processed hidden layer is determined as the encoded time-series pulse feature.

13. The high-precision ultrasonic testing method for internal micro-defects in bearing steel balls according to claim 1, characterized in that, The decoding of timing pulse characteristics to determine the absolute acoustic time of flight of internal micro-defects includes: The pulse timing features output by the spiking neural network model are analyzed to identify the pulse clusters corresponding to each micro-defect echo and extract the start emission time of each pulse cluster. The initial emission time of each pulse cluster is mapped back to the time axis coordinate system of the original time domain amplitude sequence. The complete time interval from the moment the ultrasonic probe emits a pulse to the moment the micro-defect scattering echo reaches the probe receiving surface is calculated. The complete time interval is determined as the absolute acoustic flight time of the ultrasonic wave from the probe through the coupling medium into the steel ball to the micro-defect location and back.

14. The high-precision ultrasonic testing method for internal micro-defects in bearing steel balls according to claim 13, characterized in that, The method of calculating the precise spatial coordinates of micro-defects by combining the incident spatial angle of the ultrasonic probe and outputting the ultrasonic detection results of micro-defects inside the bearing steel ball includes: The arrival time of the front surface echo is extracted from the full-depth ultrasonic reflection signal to determine the one-way propagation time of the ultrasonic wave in the coupling medium; Based on the determined absolute acoustic flight time, the two-way propagation time of the ultrasonic wave in the coupling medium is subtracted, and the remaining flight time is converted into the radial depth distance of the micro-defect relative to the incident surface by combining the sound velocity inside the steel ball. Obtain the incident spatial angle parameters of the ultrasonic probe at the current scanning orientation, and combine them with the radial depth distance to locate the micro-defect to the precise spatial coordinates of the three-dimensional coordinate system inside the steel ball; The spatial coordinates and echo characteristics of micro-defects detected from each scanning orientation are summarized to generate and output the ultrasonic detection results of micro-defects inside the bearing steel ball.

15. A high-precision ultrasonic testing system for internal micro-defects in bearing steel balls, used to implement the high-precision ultrasonic testing method for internal micro-defects in bearing steel balls according to any one of claims 1 to 14, characterized in that, include: The system includes an ultrasonic signal acquisition module, a reverberation matrix generation module, an echo envelope extraction module, a pulse feature encoding module, and a defect location output module. The ultrasonic signal acquisition module acquires the full-depth ultrasonic reflection signal of the bearing steel ball and records the corresponding incident spatial angle. It digitizes the full-depth ultrasonic reflection signal into a time-domain amplitude sequence containing internal defect echoes and multiple reverberation echoes from the spherical boundary, and marks the time axis coordinates. The reverberation matrix generation module, based on the geometric diameter parameters and internal sound velocity of the steel ball, constructs a path tracing logic for multiple reflections at the spherical boundary, calculates the transit time of the ultrasonic wave propagating back and forth within the spherical boundary, and generates a temporal distribution matrix of theoretical reverberation nodes. The echo envelope extraction module dynamically aligns the temporal distribution matrix of the theoretical reverberation node with the time-domain amplitude sequence, removes the boundary reverberation components from the time-domain amplitude sequence, and extracts candidate echo envelopes containing internal micro-defect features. The pulse feature encoding module inputs the candidate echo envelope into the pulse neural network model that integrates the adaptive threshold membrane potential mechanism, encodes the temporal pulse features of the candidate echo envelope, and suppresses residual high-frequency electrical noise to obtain a pure internal micro-defect echo pulse sequence. The defect location output module decodes the timing pulse characteristics to determine the absolute acoustic time of flight of the internal micro-defect, calculates the precise spatial coordinates of the micro-defect by combining the incident spatial angle of the ultrasonic probe, and outputs the ultrasonic detection results of the internal micro-defect of the bearing steel ball.

Images