Quality detection method and system for precision hardware shaft

Abstract

The invention discloses a quality detection method and system for a precision hardware shaft, and aims to solve the problems of missing detection and existence of a blind area caused by detection parameter solidification and dynamic mismatch. Comprising the following steps: determining and generating an initial excitation signal according to a corresponding relation between defect types and frequencies; in the rotating process of the hardware shaft, synchronous and angle-by-angle amplitude and phase calibration is carried out on the excitation signal by using the real-time speed and the angle position obtained by the encoder so as to establish a stable detection base line; the detection process is optimized in real time through three layers of closed-loop feedback: first feedback: when abnormal response is monitored, the frequency is finely tuned to enhance the detection capability; second feedback, performing local adaptive adjustment on the frequency according to the geometric position of the shaft body; and third feedback, analyzing data coverage after detection is finished, and filling a detection blind area by adjusting scanning parameters. Through multiple feedback and dynamic calibration, comprehensive, high-precision and non-blind area self-adaptive detection of the defects of the rotating hardware shaft is realized, and the detection reliability and efficiency are improved.

Description

Technical Field

[0001] This invention relates to the field of nondestructive testing technology, and in particular to a method and system for quality inspection of precision hardware shafts. Background Technology

[0002] Precision metal shafts are key components of mechanical equipment, and their quality directly affects the overall performance and reliability of the equipment. Therefore, detecting internal defects in metal shafts is crucial during the manufacturing process. Currently, metal shaft quality inspection methods mainly rely on physical principles such as electromagnetic induction, ultrasound, or radiation. However, existing technologies have limitations in terms of inspection efficiency and accuracy.

[0003] For example, traditional ultrasonic testing methods rely on manual adjustment of probe position and scanning speed, resulting in low testing efficiency and susceptibility to operator subjectivity. While radiographic testing can achieve high precision, it poses radiation safety hazards and is costly. Furthermore, existing testing methods struggle to meet the detection needs of different defect types, typically requiring the selection of specific testing parameters and methods for particular defect types, leading to cumbersome testing procedures and failing to meet the demands for rapid and comprehensive testing.

[0004] Especially in dynamic inspection of rotating hardware shafts, existing inspection methods struggle to achieve precise synchronization between the excitation signal and the shaft's rotation, leading to unstable inspection results and the potential for blind spots. Therefore, improving the efficiency, accuracy, and comprehensiveness of hardware shaft quality inspection is a pressing issue for those skilled in the art. Current technologies lack research on adaptive excitation frequency adjustment strategies for hardware shaft quality inspection, making it difficult to dynamically optimize inspection parameters based on shaft condition and defect characteristics, thus affecting inspection effectiveness.

[0005] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of the present invention, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] In view of this, the present invention provides a quality inspection method and system for precision hardware shafts to solve the problems of reduced detection accuracy and reliability in the prior art when performing quality inspection on precision hardware shafts in a rotating state, due to the asynchronous excitation signal and shaft rotation, fixed detection parameters, and the existence of detection blind spots. By synchronizing and calibrating the excitation signal in real time, dynamically adjusting the excitation signal frequency, and optimizing the frequency scanning step size and angle sampling density, comprehensive and accurate detection of defects in precision hardware shafts can be achieved.

[0007] This invention provides a method for quality inspection of precision hardware shafts, comprising: Based on the correspondence between the defect type of the hardware shaft and the excitation frequency, the initial excitation frequency for detecting the defect type is determined, and an excitation signal containing the initial excitation frequency is generated. The driving excitation signal is applied to the rotating hardware shaft, and the rotation speed and real-time angular position of the hardware shaft are obtained at the same time. Based on the rotation speed and real-time angular position, the output rhythm, amplitude or phase of the excitation signal is synchronized and calibrated. The system monitors the signal response based on the synchronized and calibrated excitation signal in real time, and performs a first feedback adjustment on the frequency of the excitation signal when the characteristic value of the signal response deviates from the preset range, so as to enhance the detection capability of the signal response. The matching degree between the geometric position of the hardware shaft and the frequency of the excitation signal after the first feedback adjustment is analyzed. When there is a mismatch area between the two, the frequency of the excitation signal is adjusted by the second feedback to adapt to the detection requirements of the mismatch area. The system continuously collects defect response data based on the excitation signal adjusted by the second feedback, and adjusts the frequency scanning step size or angle sampling density of the excitation signal by the third feedback according to whether there are missing frequency bands or angle ranges in the defect response data, so as to cover the missing frequency bands or angle ranges.

[0008] In some optional embodiments, the initial excitation frequency is determined based on the correspondence between the defect type of the hardware shaft and the excitation frequency, including: From a database containing historical capacitance response data of different defect types at different excitation frequencies, the optimal excitation frequency range corresponding to the defect type to be tested is queried and determined, and the initial excitation frequency is selected from the range.

[0009] In some optional embodiments, the output rhythm, amplitude, or phase of the excitation signal is synchronized and calibrated, including: By using a phase-locked loop algorithm, the output frequency of the excitation signal is locked to an integer multiple of the rotation frequency of the hardware shaft in order to achieve synchronization of the output rhythm; For each real-time angular position, the amplitude calibration and phase calibration values ​​for that angular position are calculated using linear interpolation and then applied to the excitation signal.

[0010] In some optional embodiments, a first feedback adjustment is made to the frequency of the excitation signal, including: When the characteristic value of the signal response deviates from the preset range, the gradient descent algorithm is used to iteratively adjust the frequency value of the current excitation signal with the goal of maximizing the deviation of the characteristic value.

[0011] In some optional embodiments, a second feedback adjustment is performed on the frequency of the excitation signal, including: The natural frequencies of the mismatched regions are calculated using the finite element method, and then, combined with a genetic algorithm, the measured frequencies are adjusted to an excitation frequency that is compatible with the natural frequencies.

[0012] In some alternative embodiments, after determining the initial excitation frequency, the method further includes: The real-time response signal of the tested hardware shaft is input into a pre-trained autoencoder model for reconstructing the response signal of the defect-free hardware shaft; Calculate the reconstruction error between the real-time response signal and the model reconstruction output signal; The frequency component that generates an error greater than a preset threshold is determined as the initial excitation frequency.

[0013] In some optional embodiments, the method further includes: Extract the latent space feature vector of the real-time response signal that generates an error greater than a preset threshold in the autoencoder model.

[0014] In some optional embodiments, the method further includes: By performing unsupervised clustering on multiple latent space feature vectors, online classification of different defects can be achieved.

[0015] In some optional embodiments, the method further includes: Based on the manual labeling of the online classification results, the autoencoder model is updated using transfer learning to enable the model to recognize the labeled defect types.

[0016] In some alternative embodiments, the excitation signal is one of a capacitive field signal, an eddy current signal, or an ultrasonic signal.

[0017] This invention provides a quality inspection system for precision hardware shafts, comprising: The excitation and synchronization calibration unit is configured to: determine the initial excitation frequency according to the correspondence between the defect type of the hardware shaft and the excitation frequency, generate and drive the excitation signal containing the initial excitation frequency to act on the rotating hardware shaft, and at the same time acquire the rotation speed and real-time angular position of the hardware shaft, and synchronize and calibrate the output rhythm, amplitude or phase of the excitation signal according to the rotation speed and real-time angular position. The first feedback control unit is connected to the excitation and synchronization calibration unit and is configured to: monitor the signal response based on the synchronized and calibrated excitation signal in real time, and send a command to the excitation and synchronization calibration unit when the characteristic value of the signal response deviates from the preset range, so as to perform the first feedback adjustment on the frequency of the excitation signal; The second feedback control unit is connected to the first feedback control unit and is configured to: analyze the matching degree between the geometric position of the hardware shaft and the frequency of the excitation signal adjusted by the first feedback, and when there is a mismatch area between the two, send a command to the excitation and synchronization calibration unit to perform a second feedback adjustment on the frequency of the excitation signal; The third feedback control unit, connected to the second feedback control unit, is configured to: continuously analyze the defect response data based on the excitation signal adjusted by the second feedback, and when there are frequency bands or angle ranges with missing signals in the defect response data, issue instructions to the excitation and synchronization calibration unit to adjust the frequency scan step size or angle sampling density of the excitation signal.

[0018] In some optional embodiments, the excitation and synchronization calibration unit includes: An encoder for real-time acquisition of the angular position of hardware shafts; A processor connected to the encoder runs a phase-locked loop algorithm to synchronize the output frequency of the excitation signal with the rotation frequency of the hardware shaft, and runs a linear interpolation algorithm to calculate the amplitude and phase calibration values ​​based on the real-time angular position acquired by the encoder.

[0019] In some alternative embodiments, the second feedback control unit is further configured to: Run a finite element analysis model to calculate the natural frequencies of the mismatched region; Furthermore, a genetic algorithm is run to calculate the optimal excitation frequency for the second feedback adjustment based on the inherent frequency.

[0020] In some optional embodiments, it further includes: The intelligent diagnostic module is configured to: run a pre-trained autoencoder model, receive signal responses to calculate reconstruction errors, determine the initial excitation frequency based on the reconstruction errors, and extract the latent space feature vectors of the signal responses for online classification or identification of defects.

[0021] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit the invention.

[0022] The quality inspection method and system for precision hardware shafts of the present invention have the following beneficial effects: This invention establishes a correspondence between defect types and excitation frequencies to achieve precise selection of the initial excitation frequency. A synchronization and calibration mechanism ensures the synchronization of the excitation signal and shaft rotation, improving detection accuracy. A multi-level feedback adjustment mechanism optimizes the excitation signal frequency, enhancing the detection capability of defect signals and adapting to the detection needs of different geometric positions. By adjusting the frequency scanning step size and angle sampling density, detection blind spots are effectively covered, achieving comprehensive detection and improving the accuracy and reliability of hardware shaft quality inspection. Attached Figure Description

[0023] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings.

[0024] Figure 1 This is a flowchart of a quality inspection method for precision hardware shafts according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a precision hardware shaft quality inspection system according to an embodiment of the present invention. Detailed Implementation

[0025] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the invention will be more comprehensive and complete, and will fully convey the concept of the exemplary embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0026] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0027] The flowchart shown in the attached diagram is merely an illustrative example and does not necessarily include all steps. For example, some steps may be broken down, while others may be combined or partially combined. Therefore, the actual execution order may change depending on the specific circumstances.

[0028] This invention relates to the field of quality inspection of precision metal shafts. It utilizes precise synchronization of the excitation signal and the shaft's rotational state, along with adaptive frequency adjustment, to achieve comprehensive detection of potential defects. The techniques relied upon include: excitation signal generation and control technology, which uses a signal of a specific frequency to excite different types of defects within the shaft to generate detectable responses; precise measurement technology of the shaft's angular position, which uses encoders and other devices to acquire the shaft's rotational angle in real time, providing a basis for signal synchronization; and signal processing and analysis technology, used to extract defect response characteristics and adjust the frequency and phase of the excitation signal based on these characteristics. Through these technologies, this invention achieves comprehensive quality inspection of metal shafts under dynamic rotation conditions, improving the accuracy and reliability of the inspection. This invention combines multiple technologies and optimizes the matching degree between the excitation signal and the shaft's state through a multi-level feedback adjustment mechanism, ultimately achieving accurate identification and location of defects in the metal shaft.

[0029] like Figure 1 As shown, this invention provides a quality inspection method for precision metal shafts. Through a multi-stage, adaptive excitation signal control mechanism, it aims to solve the problems of poor adaptability to complex defect types and the tendency to generate blind spots in existing metal shaft quality inspection methods, thereby improving the comprehensiveness and reliability of the inspection. The method includes: S100. Based on the correspondence between the defect type and the excitation frequency of the hardware shaft, the initial excitation frequency for detecting the defect type is determined, and an excitation signal containing the initial excitation frequency is generated. This embodiment of the invention relies on a pre-established database or model relating defect types to excitation frequencies. This database or model characterizes the unique capacitance, eddy current, and ultrasonic response characteristics exhibited by specific types of defects, such as cracks, porosity, and inclusions, at different excitation frequencies. By querying this database or model, the optimal excitation frequency range for one or more defects to be detected can be determined, and one or more initial excitation frequencies can be selected to generate the corresponding excitation signal. Specifically, lookup tables, regression models, neural networks, and other methods can be used to construct the correspondence between defect types and excitation frequencies. For example, the Support Vector Machine (SVM) algorithm can be used to classify the optimal excitation frequency ranges corresponding to different defect types in historical detection data, establishing a mapping relationship between defect types and frequency ranges. The generation of the excitation signal can employ Direct Digital Frequency Synthesis (DDS) technology to generate signals such as sine waves, square waves, and pulse waves containing the selected initial excitation frequency, or to generate composite signals containing multiple frequency components. This step ensures targeted detection, effectively stimulating the response of the target defect and laying the foundation for subsequent accurate detection. As an alternative implementation, the determination of the initial excitation frequency is not limited to database searching. It can also be achieved by performing a preliminary frequency scan of the tested metal shaft, analyzing its response spectrum, and selecting the frequency with the highest response amplitude as the initial excitation frequency. For example, a frequency range, such as 1kHz to 10kHz, can be preset, with the excitation frequency increasing in steps such as 100Hz. The capacitance response of the metal shaft can be monitored in real time, and the frequency corresponding to the peak capacitance response can be selected as the initial excitation frequency.

[0030] S200: The drive excitation signal is applied to the rotating hardware shaft, and the rotation speed and real-time angular position of the hardware shaft are obtained. Based on the rotation speed and real-time angular position, the output rhythm, amplitude or phase of the excitation signal is synchronized and calibrated.

[0031] In this embodiment of the invention, a generated excitation signal is applied to a rotating hardware shaft, while sensors such as encoders and tachometers monitor the shaft's rotational speed and angular position in real time. Based on the acquired rotational information, the output rhythm, amplitude, or phase of the excitation signal is precisely synchronized and calibrated to ensure that the excitation signal and the hardware shaft's rotational motion are coordinated, avoiding detection errors or blind spots caused by asynchrony or signal drift. Specifically, synchronization can be achieved using phase-locked loop (PLL) technology, locking the frequency of the excitation signal to an integer multiple of the hardware shaft's rotational frequency, ensuring that the output rhythm of the excitation signal is strictly synchronized with the hardware shaft's rotational period. For example, if the hardware shaft's rotational frequency is 10Hz, the excitation signal frequency can be locked to 50Hz or 100Hz to achieve frequency multiplication excitation. Calibration can be achieved through table lookup or real-time calculation, adjusting the amplitude or phase of the excitation signal according to the hardware shaft's real-time angular position to compensate for signal attenuation or distortion caused by factors such as shaft structure and material inhomogeneity. For example, the signal response characteristics of the hardware shaft at different angular positions can be pre-calibrated, and a mapping table between angular position and amplitude / phase calibration value can be established. During the detection process, calibration is performed by looking up the table based on the real-time angular position. As an alternative implementation method, synchronization is not limited to PLL technology; a timestamp synchronization method can also be used. That is, timestamps are recorded during the output of the excitation signal and the acquisition of the angular position. By comparing the timestamps, the output time of the excitation signal can be precisely adjusted to achieve time synchronization.

[0032] S300 monitors the signal response based on the synchronized and calibrated excitation signal in real time, and performs a first feedback adjustment on the frequency of the excitation signal when the characteristic value of the signal response deviates from the preset range, so as to enhance the detection capability of the signal response.

[0033] In this embodiment of the invention, after the excitation signal is synchronized and calibrated, the signal response of the hardware shaft is monitored in real time. Feature values ​​of the response signal, such as amplitude, phase, and frequency, are extracted and compared with a preset normal range. When the feature value deviates from the preset range, it indicates a possible defect or other abnormality. At this point, a first feedback adjustment of the excitation signal frequency is needed to optimize the excitation conditions, improve the detection capability of the signal response, and ensure that the feature information of the defect can be effectively captured. Specifically, feature value extraction can employ digital signal processing (DSP) techniques, such as Fast Fourier Transform (FFT) and wavelet transform, to convert the time-domain signal into a frequency-domain signal and extract feature parameters such as amplitude, phase, and frequency. The preset normal range can be determined through statistical analysis of historical data or by establishing a simulation model. For example, the signal response feature values ​​of a defect-free hardware shaft at different excitation frequencies can be statistically analyzed, and their mean and standard deviation can be calculated. The mean ± 3 times the standard deviation can be used as the normal range. The first feedback adjustment can employ optimization algorithms such as gradient descent or simulated annealing to iteratively adjust the frequency of the excitation signal with the goal of maximizing the signal response feature value or minimizing the feature value deviation. For example, the gradient descent method can be used, with the signal response amplitude as the objective function, to calculate the gradient of the amplitude with respect to frequency, and adjust the frequency according to the gradient direction until the amplitude reaches its maximum value or converges. As an alternative implementation, the first feedback adjustment is not limited to adjusting the frequency; it can also adjust the amplitude, phase, or waveform of the excitation signal to optimize the excitation conditions and improve the defect detection rate.

[0034] S400. Analyze the matching degree between the geometric position of the hardware shaft and the frequency of the excitation signal after the first feedback adjustment, and when there is a mismatch area between the two, perform a second feedback adjustment on the frequency of the excitation signal to adapt to the detection requirements of the mismatch area.

[0035] In this embodiment of the invention, after the first feedback adjustment, the matching degree between the geometric position of the hardware shaft and the current excitation signal frequency is further analyzed. Considering that factors such as the shape, size, and material of the hardware shaft will affect its frequency response characteristics at different positions, a second feedback adjustment is needed for areas where the geometric position and frequency configuration do not match. This ensures that the frequency of the excitation signal matches the natural frequency or resonant frequency of that area, thereby improving detection sensitivity. Specifically, a finite element analysis (FEA) method can be used to establish a geometric model of the hardware shaft and calculate its natural frequencies at different positions. For example, finite element analysis software such as ANSYS and COMSOL can be used to perform modal analysis on the hardware shaft to obtain the distribution of its first few natural frequencies. The second feedback adjustment can select a frequency close to the natural frequency of the mismatched area as the new excitation frequency based on the results of the finite element analysis. For example, if the natural frequency in a certain area is 150Hz, and the current excitation frequency is 100Hz, the excitation frequency can be adjusted to 150Hz. As an alternative implementation, the second feedback adjustment is not limited to adjusting the frequency; it can also adjust the excitation position or excitation direction of the excitation signal to optimize the excitation effect and improve the detection coverage.

[0036] S500 continuously collects defect response data based on the excitation signal adjusted by the second feedback, and adjusts the frequency scanning step size or angle sampling density of the excitation signal by the third feedback according to whether there are missing frequency bands or angle ranges in the defect response data, so as to cover the missing frequency bands or angle ranges.

[0037] In this embodiment of the invention, after the second feedback adjustment, defect response data of the hardware shaft is continuously collected, and the data is analyzed to determine if there are frequency bands or angle ranges with missing signals. These areas with missing signals may represent detection blind spots or areas with low sensitivity. In this case, a third feedback adjustment is needed to adjust the frequency scan step size or angle sampling density of the excitation signal to ensure comprehensive coverage of all areas of the hardware shaft, thereby improving the integrity and reliability of the detection. Specifically, the determination of frequency bands or angle ranges with missing signals can be achieved through spectrum analysis and statistical analysis. For example, a Fourier transform can be performed on the collected defect response data to analyze its spectrum distribution. If the signal energy of a certain frequency band is found to be significantly lower than that of other frequency bands, it can be considered that there is a signal missing in that frequency band. The third feedback adjustment can reduce the frequency scan step size or increase the angle sampling density based on the signal missing situation. For example, if a signal missing is found in the 50Hz to 60Hz frequency band, the frequency scan step size can be reduced from 1Hz to 0.5Hz to scan that frequency band more precisely. As an alternative implementation, the third feedback adjustment is not limited to adjusting the frequency scanning step size or the angle sampling density, but can also adjust the excitation intensity or excitation time of the excitation signal to improve the signal-to-noise ratio and improve the detection effect.

[0038] In summary, this invention, through a multi-stage, adaptive excitation signal control mechanism, achieves comprehensive and accurate detection of internal defects in metal shafts, effectively solving the problems of poor defect adaptability and blind spots in existing technologies. The method determines the initial excitation frequency by establishing a pre-defined correspondence between defect types and excitation frequencies. Then, using phase-locked loop (PLL) technology and lookup table calibration methods, the excitation signal is synchronized and calibrated with the rotational motion of the metal shaft. Next, a gradient descent algorithm is used to perform a first feedback adjustment on the excitation signal frequency where the signal response characteristic value deviates from the preset range. Then, finite element analysis and genetic algorithms are used for a second feedback adjustment on areas where the geometric position and frequency configuration do not match. Finally, a third feedback adjustment is performed on missing frequency bands or angle intervals by reducing the frequency scanning step size or increasing the angle sampling density, thereby achieving comprehensive detection coverage. This method can effectively detect internal defects such as cracks, porosity, and inclusions in metal shafts, improving detection accuracy and reliability, and has significant industrial application value.

[0039] In some embodiments, the initial excitation frequency is determined based on the correspondence between the defect type of the hardware shaft and the excitation frequency, including: From a database containing historical capacitance response data of different defect types at different excitation frequencies, the optimal excitation frequency range corresponding to the defect type to be tested is queried and determined, and the initial excitation frequency is selected from the range.

[0040] This invention aims to address the problem of selecting the optimal excitation frequency for specific defects to improve detection targeting. Different defects exhibit frequency-dependent sensitivity to excitation signals; an efficient frequency selection based on prior knowledge can be achieved through a pre-established database. As a specific implementation, the database is constructed through experiments or simulations, storing eddy current or ultrasonic response data of defects such as cracks and porosity at different frequencies. For example, to detect surface cracks, the optimal response range can be found in the database to be 10kHz-20kHz, and 15kHz can be selected as the initial excitation frequency.

[0041] To further optimize, a small-range, high-density frequency scan can be performed within the frequency range defined in the database, and the signal-to-noise ratio (SNR) of the response signal can be monitored in real time. The frequency with the highest SNR is then selected as the final initial excitation frequency. Furthermore, to simultaneously detect multiple defects, several optimal frequencies corresponding to different defect types can be selected from the database and combined into a multi-frequency composite excitation signal using optimization techniques such as genetic algorithms, thereby improving the comprehensiveness of the detection.

[0042] In other embodiments, the synchronization and calibration of the excitation signal includes: using a phase-locked loop (PLL) algorithm to precisely lock the output frequency of the excitation signal to an integer multiple of the rotation frequency of the hardware shaft to achieve synchronization of the output rhythm; at the same time, for each real-time angular position fed back by the encoder, the corresponding amplitude and phase calibration amount is calculated and applied by linear interpolation to compensate for systematic signal fluctuations introduced by factors such as installation eccentricity or geometric tolerances, and to ensure the stability and purity of the detection baseline.

[0043] In some embodiments, synchronizing and calibrating the output rhythm, amplitude, or phase of the excitation signal includes: By using a phase-locked loop algorithm, the output frequency of the excitation signal is locked to an integer multiple of the rotation frequency of the hardware shaft in order to achieve synchronization of the output rhythm; For each real-time angular position, the amplitude calibration and phase calibration values ​​for that angular position are calculated using linear interpolation and then applied to the excitation signal.

[0044] In this embodiment of the invention, a digital phase-locked loop (DPLL) or an analog phase-locked loop (APLL) can be used to achieve synchronization. DPLL offers high flexibility, allowing for programmable adjustment of loop parameters to optimize performance; APLL has a simple structure and fast response. During synchronization, the excitation frequency can be locked to an integer multiple of the rotation frequency to improve the signal-to-noise ratio of specific frequency components.

[0045] To achieve calibration, in addition to linear interpolation, polynomial interpolation or spline interpolation can also be used to strike a balance between accuracy and smoothness. Specifically, a standard axis can be scanned in advance to create a calibration lookup table containing the amplitude and phase deviations at each angular position. During testing, the calibration value is retrieved from this table based on the real-time angle feedback from the encoder and applied to the excitation signal. To cope with environmental changes, adaptive calibration methods can also be used, such as using a Kalman filter algorithm to estimate and update calibration parameters in real time.

[0046] In some embodiments, the frequency of the excitation signal is adjusted by a first feedback adjustment, including: When the characteristic value of the signal response deviates from the preset range, the gradient descent algorithm is used to iteratively adjust the frequency value of the current excitation signal with the goal of maximizing the deviation of the characteristic value.

[0047] This invention aims to automatically optimize the excitation frequency to obtain the clearest defect signal when a suspected defect is detected. Here, "characteristic value" refers to a signal parameter that can characterize the presence of a defect, such as the response amplitude of a capacitive sensor, the impedance value of an eddy current detector, or the phase angle of an ultrasonic signal. When this characteristic value exceeds a preset normal fluctuation range, feedback adjustment is triggered.

[0048] Specifically, taking capacitance detection as an example, if the response amplitude suddenly drops below a threshold, the system determines that an anomaly exists. At this time, the first feedback control unit initiates the gradient descent algorithm. The objective function of the algorithm is set to maximize the difference between the normal amplitude and the current abnormal amplitude, i.e., the deviation. Based on the response amplitude at the current excitation frequency, the algorithm calculates the gradient direction of the objective function with respect to frequency, and adjusts the excitation frequency along this direction in a set step size, such as 0.1 kHz. This process is implemented by a digital signal processor (DSP) controlling a signal generator, and it is iterated repeatedly. After each iteration, the system evaluates the new response amplitude and continues to adjust the frequency until the deviation converges to its maximum value, or reaches the preset number of iterations. The frequency at this point is the frequency with the strongest detection capability for this specific defect.

[0049] Besides gradient descent, other optimization algorithms such as conjugate gradient or Newton's method can also be used. This real-time, adaptive frequency "focusing" mechanism significantly enhances the ability to capture weak defect signals, thereby greatly improving detection sensitivity and reliability.

[0050] In some embodiments, a second feedback adjustment is performed on the frequency of the excitation signal, including: The natural frequencies of the mismatched regions are calculated using the finite element method, and then, combined with a genetic algorithm, the measured frequencies are adjusted to an excitation frequency that is compatible with the natural frequencies.

[0051] The present invention aims to solve the problem of local detection adaptability caused by abrupt changes in the geometry of hardware shafts, such as steps, keyways, and through holes. These geometrically discontinuous regions, or "mismatched regions," have different response characteristics to excitation signals than the smooth parts of the shaft. If a uniform frequency is still used for detection, it may lead to a decrease in sensitivity.

[0052] Specifically, the system first uses a CAD model of the hardware shaft and then employs FEA software (such as ANSYS or ABAQUS) to model and simulate these mismatched regions. By inputting parameters such as the Young's modulus and Poisson's ratio of the material, FEA can accurately calculate the local resonant frequencies, i.e., natural frequencies, of these regions under specific excitation methods (such as capacitive fields or eddy current fields). These natural frequencies are the frequencies at which the region is most easily excited to produce a significant response.

[0053] Then, when the encoder instructs the detection probe to move into the mismatch area, the second feedback control unit is activated. It initiates a genetic algorithm, using the natural frequency calculated by the FEA as the optimization objective. The genetic algorithm uses a frequency range, such as ±10% of the natural frequency, as its search space, with the objective function being to minimize the deviation between the actual excitation frequency and the natural frequency, or to maximize the expected signal response intensity in that region. Through multiple generations of selection, crossover, and mutation operations, the genetic algorithm can quickly converge and output an optimal excitation frequency highly adapted to the natural frequency. The system then instructs the signal generator to switch to this frequency until the probe leaves the area.

[0054] Alternatively, other global optimization algorithms such as Particle Swarm Optimization (PSO) can be used. This combination of simulation and intelligent optimization ensures that the detection of any complex geometric position on the hardware axis is performed at the optimal frequency, effectively improving the comprehensiveness and reliability of the detection.

[0055] In some embodiments, after determining the initial excitation frequency, the method further includes: The real-time response signal of the tested hardware shaft is input into a pre-trained autoencoder model for reconstructing the response signal of the defect-free hardware shaft; Calculate the reconstruction error between the real-time response signal and the model reconstruction output signal; The frequency component that generates an error greater than a preset threshold is determined as the initial excitation frequency.

[0056] This invention aims to select the initial excitation frequency more intelligently and accurately through artificial intelligence. An autoencoder model is used to learn and establish a digital benchmark for the response signal of a "healthy" hardware shaft. This model, such as a convolutional autoencoder (CAE), is trained on a large amount of response data from defect-free samples, which can come from actual measurements or simulations, enabling it to accurately reconstruct any signal with a "healthy" input.

[0057] In actual testing, a wideband initial probe signal is first applied to the shaft under test, and its response is acquired. This response signal is then input into a trained autoencoder. If the shaft is defect-free, its response signal conforms to the model's "healthy" perception, and the model can reproduce it with minimal reconstruction error. Conversely, if the shaft has defects, the defects will disrupt the signal pattern, causing the response signal to deviate from the "healthy" baseline. When the model attempts to reconstruct this "abnormal" signal, due to a lack of understanding of the defect pattern, its output will differ significantly from the original input, resulting in a large reconstruction error.

[0058] By analyzing the distribution of reconstruction errors in the frequency domain, the system identifies frequency points where error values ​​exceed a preset statistical threshold, such as the mean plus three standard deviations. These frequency points are precisely those where signal distortion is most severe due to the presence of defects, i.e., the frequencies most sensitive to these defects. The system then determines these high-error frequencies as the optimal initial excitation frequencies for subsequent fine-tuning detection, thereby significantly improving the targeting of frequency selection and detection efficiency.

[0059] In some embodiments, the method further includes: Extract the latent space feature vector of the real-time response signal that generates an error greater than a preset threshold in the autoencoder model.

[0060] In this embodiment of the invention, the aim is to extract highly condensed and discriminative defect fingerprint information from anomalous signals for deeper analysis. The autoencoder model consists of an encoder and a decoder. The encoder is responsible for compressing the high-dimensional input signal into a low-dimensional latent space representation, i.e., a latent space feature vector.

[0061] In practice, when the system determines that a response signal is abnormal by calculating the reconstruction error, it does not discard the signal. Instead, it passes it through the encoder part of the model again, directly reading and saving the low-dimensional feature vector output by the encoder, such as a 16-dimensional or 32-dimensional numerical vector. This vector is a highly condensed mathematical expression of the original abnormal signal within the model. It filters out noise and redundant information, preserving the key structured features caused by the defect to the greatest extent possible.

[0062] These extracted latent space feature vectors, due to their low dimensionality and high information density, are very suitable as inputs for subsequent advanced analysis tasks (such as automatic defect classification, severity assessment, or similar defect retrieval), providing a key data foundation for achieving intelligent upgrades from "detection" to "diagnosis".

[0063] In some embodiments, the method further includes: By performing unsupervised clustering on multiple latent space feature vectors, online classification of different defects can be achieved.

[0064] This invention provides the system with the ability to automatically detect and distinguish different defect types without any prior knowledge of defects. For defects of the same type, even if their specific response signals differ, the latent space feature vectors extracted by the autoencoder will be relatively clustered in the feature space. Therefore, clustering algorithms can be used to automatically cluster these vectors.

[0065] In practice, the system collects the latent space feature vectors corresponding to all signals identified as abnormal within a production batch or a set time period, forming a dataset. Then, the system initiates an unsupervised clustering algorithm, such as DBSCAN (Density-Based Spatial Clustering). The advantage of choosing DBSCAN is that it does not require pre-specifying the number of clusters K, and it can identify clusters of arbitrary shapes while effectively identifying outliers as noise. This makes it very suitable for handling complex and diverse defect data in real industrial environments.

[0066] The algorithm processes the dataset, dividing the feature vectors into several independent clusters. For example, the system might automatically output three clusters: C1, C2, and C3. This means that the system has automatically categorized all detected unknown defects into three types at the data level. Each cluster then represents a potential, unidentified defect type. Subsequently, technicians only need to extract one or two samples from each cluster for physical analysis, such as slicing, to name and characterize the entire defect category, such as C1 representing surface cracks and C2 representing internal porosity. This online automatic classification greatly reduces the burden of manual analysis and is a key step in transitioning from automated detection to intelligent diagnosis.

[0067] In some embodiments, the method further includes: Based on the manual labeling of the online classification results, the autoencoder model is updated using transfer learning to enable the model to recognize the labeled defect types.

[0068] In this embodiment of the invention, the method is endowed with the ability to self-evolve, efficiently transforming human experience into model intelligence. The unknown defect categories discovered by the aforementioned unsupervised clustering are transformed into labeled training data through small-batch manual verification, and this data is used to quickly and cost-effectively upgrade existing models.

[0069] In practice, technicians perform sampling physical analysis on each cluster generated by clustering, such as C1 and C2, to confirm that C1 is a "surface crack" and C2 is an "internal pore". This confirmation and naming process is called "manual labeling". The system collects these latent space feature vectors with real labels such as "crack" and "pore" to form a small, high-quality labeled dataset.

[0070] The system then initiates a transfer learning process. Instead of training the entire massive autoencoder from scratch, it employs a more efficient strategy: Freeze the feature extraction layer: Treat the encoder part of the original autoencoder model as a mature and general feature extractor, and freeze the weight parameters of all its network layers so that they are no longer updated.

[0071] Add and train a classification head: At the end of the frozen encoder, add a brand new, small network module specifically for classification, such as a "classification head" consisting of one or two fully connected layers and a Softmax activation function.

[0072] Fine-tuning: This involves training the newly added "classification head" using only the aforementioned small labeled dataset. Due to the small amount of training data and the limited number of network layers, this process typically takes only a few minutes.

[0073] After training, the system has an upgraded model. It not only retains the original ability to detect unknown anomalies, such as through error reconstruction, but also gains the ability to directly identify and output calibrated defect types such as "surface cracks" and "internal porosity", achieving a qualitative leap from automated detection to intelligent diagnosis.

[0074] In some embodiments, the excitation signal is one of a capacitive field signal, an eddy current signal, or an ultrasonic signal.

[0075] In this embodiment of the invention, dynamic adaptive detection based on multiple closed-loop feedback can be flexibly applied to various mainstream non-contact non-destructive testing physical principles, and has wide applicability. The specific excitation signal selected depends on the material of the metal shaft being tested, the expected defect type, and the required detection depth.

[0076] When using a capacitive field signal: The system is configured with a ring or array-type capacitive sensor. An excitation signal generator applies an alternating voltage with a frequency ranging from 1 kHz to 1 MHz to the sensor. The system monitors minute changes in capacitance in real time using a high-precision capacitance measurement circuit, such as an LCR meter or a differential capacitance bridge. This method is extremely sensitive to surface cracks, scratches, and corrosion, enabling the detection of micron-level surface defects.

[0077] When using eddy current signals: The system is configured with one or more eddy current probes, which can be differential or absolute. An excitation signal generator applies an alternating current with a frequency in the range of 10kHz to 10MHz to the probe coil. The system analyzes the disturbances in the eddy current field, typically several millimeters deep, on and near the surface of the metal shaft by detecting changes in the probe coil impedance. This method is highly suitable for conductive materials and has excellent detection capabilities for near-surface cracks, pores, and inclusions.

[0078] When using ultrasonic signals: The system is equipped with one or more ultrasonic transducers, such as probes, employing pulse-echo or phased array technology. An excitation signal generator drives the transducers to emit ultrasonic pulses with frequencies ranging from 5MHz to 50MHz into the hardware shaft. The system locates and assesses deep internal defects, such as internal cracks, porosity, or delamination, by analyzing the flight time, amplitude, and phase of the received reflected or transmitted echo signals.

[0079] Regardless of the excitation method used, the synchronous calibration and three-layer feedback adjustment mechanism proposed in this invention can effectively apply to the corresponding signal processing flow to achieve comprehensive optimization of detection performance.

[0080] like Figure 2 As shown, this embodiment of the invention provides a quality inspection system for precision hardware shafts, including an excitation and synchronization calibration unit M100, a first feedback control unit M200, a second feedback control unit M300, and a third feedback control unit M400.

[0081] The excitation and synchronization calibration unit M100 generates an excitation signal and precisely synchronizes and calibrates it with the rotational state of the hardware shaft. The excitation signal is an energy signal used to stimulate internal defects in the hardware shaft to produce a detectable response; this signal can be electromagnetic, ultrasonic, etc. The initial excitation frequency is a pre-set frequency value based on the material properties and potential defect types of the hardware shaft, used to initially stimulate the defect response. This unit determines the initial excitation frequency based on the correspondence between the defect type of the hardware shaft and the excitation frequency, enabling targeted detection and avoiding the inefficiency of blindly scanning the frequency range. It generates and drives an excitation signal containing the initial excitation frequency onto the rotating hardware shaft to achieve preliminary detection. Rotational speed and real-time angular position are key parameters describing the motion state of the hardware shaft; obtaining these provides a basis for subsequent synchronization and calibration. Furthermore, by synchronizing and calibrating the output rhythm, amplitude, or phase of the excitation signal based on the rotational speed and real-time angular position, it ensures that the excitation signal and the rotational motion of the hardware shaft are coordinated, improving the accuracy and reliability of the detection. The excitation signal can be specifically selected from one or more combinations of capacitive field signals, eddy current signals, or ultrasonic signals. Specifically, capacitive field signals are suitable for detecting micro-cracks on the surface, eddy current signals are sensitive to internal defects in metals, and ultrasonic signals can penetrate deeper into the material for detection.

[0082] The first feedback control unit M200 monitors the signal response generated after the excitation signal is applied to the hardware shaft in real time and determines whether there is an anomaly based on the response characteristics. The characteristic values ​​of the signal response refer to electrical or acoustic parameters that reflect the internal defect state of the hardware shaft, such as capacitance, impedance, and sound pressure. This unit is connected to the excitation and synchronization calibration unit M100 to ensure real-time reception and processing of signal response data. Real-time monitoring of the signal response based on the synchronized and calibrated excitation signal is a key step in defect detection; continuous observation of changes in the response signal can promptly identify potential defects. When the characteristic values ​​of the signal response deviate from the preset range, it indicates a potential defect, requiring adjustment of the excitation signal. Sending commands to the excitation and synchronization calibration unit M100 to perform first feedback adjustment of the excitation signal frequency achieves closed-loop control of the detection process. Adjusting the frequency of the excitation signal optimizes the detection effect and improves detection sensitivity. More specifically, this unit can incorporate a pre-trained autoencoder model. This model is used to reconstruct the response signal of the defect-free hardware shaft. By comparing the reconstruction error between the real-time response signal and the model's reconstructed output signal, the anomaly of the signal response can be determined more accurately. The autoencoder model can be updated using transfer learning methods to enable it to identify identified defect types.

[0083] The second feedback control unit M300 analyzes the matching degree between the geometric position of the hardware shaft and the frequency of the excitation signal adjusted by the first feedback, and performs a secondary frequency adjustment when a mismatch exists. Here, the geometric position of the hardware shaft refers to the physical structural characteristics of the shaft, such as diameter, length, and shape, which affect signal propagation and response. The matching degree of the excitation signal frequency refers to whether the excitation frequency matches the response characteristics of a specific geometric position; mismatch may lead to poor detection results. When a mismatch exists, it indicates that the current frequency configuration cannot effectively detect defects in that area and adjustment is required. A command is sent to the excitation and synchronization calibration unit M100 to perform a second feedback adjustment of the excitation signal frequency to optimize the frequency configuration and improve the detection capability for defects in specific areas. In a preferred embodiment, this unit can run a finite element analysis model to calculate the inherent frequency of the mismatch area and, combined with a genetic algorithm, adjust the measured frequency to an excitation frequency compatible with the inherent frequency, thereby improving detection efficiency and accuracy.

[0084] The third feedback control unit M400 continuously analyzes the defect response data of the excitation signal adjusted by the second feedback, and adjusts the frequency scan step size or angle sampling density according to the signal missing information in the data to achieve comprehensive detection coverage. The missing frequency bands or angle ranges refer to certain frequency or angle ranges where the detection signal strength is significantly weakened or even disappears, which may be due to defect location, signal attenuation, or other reasons. Adjusting the frequency scan step size or angle sampling density of the excitation signal can increase the detection probability of blind zones, improving the comprehensiveness and reliability of the detection. More specifically, this unit can extract defect response features based on a high-density sampling dataset using an adaptive sampling method, and employ principal component analysis to obtain comprehensive detection data, thereby more accurately determining the defect state of the shaft rotation.

[0085] In summary, this embodiment of the invention generates and synchronously calibrates an excitation signal through an excitation and synchronization calibration unit M100, and performs multi-level feedback adjustments on the frequency, scan step size, and angle sampling density of the excitation signal through a first feedback control unit M200, a second feedback control unit M300, and a third feedback control unit M400, thereby achieving comprehensive and accurate detection of defects in hardware shafts. This multi-level feedback adjustment mechanism can effectively cope with various complex situations, improve the accuracy and reliability of detection, reduce the false negative rate, and thus improve the product quality and service life of hardware shafts.

[0086] In some embodiments, the excitation and synchronization calibration unit M100 includes: An encoder for real-time acquisition of the angular position of hardware shafts; A processor connected to the encoder runs a phase-locked loop algorithm to synchronize the output frequency of the excitation signal with the rotation frequency of the hardware shaft, and runs a linear interpolation algorithm to calculate the amplitude and phase calibration values ​​based on the real-time angular position acquired by the encoder.

[0087] In this embodiment of the invention, the encoder, which can be absolute or incremental, provides a high-precision angular displacement reference. The processor, such as a DSP or FPGA, receives the pulses or digital codes output by the encoder and calculates the instantaneous rotational frequency of the shaft in real time. This frequency serves as the reference input for the PLL algorithm, which in turn controls the clock of the excitation signal generator, ensuring that its output frequency is precisely locked to a preset integer multiple of the rotational frequency, thereby achieving clock synchronization.

[0088] Simultaneously, the processor queries a pre-established calibration lookup table (LUT) for each precise angle value provided by the encoder. This table stores the signal amplitude and phase deviation at each angular position under standard conditions. The processor uses a linear interpolation algorithm to calculate the amplitude compensation coefficient and phase compensation value precisely corresponding to the current angle and applies them to the excitation signal to be output in real time. This precise angle-by-angle calibration effectively eliminates systematic errors caused by installation eccentricity, geometric tolerances, etc., ensuring the flatness and stability of the detection baseline.

[0089] In some embodiments, the second feedback control unit M300 is further configured to: Run a finite element analysis model to calculate the natural frequencies of the mismatched region; Furthermore, a genetic algorithm is run to calculate the optimal excitation frequency for the second feedback adjustment based on the inherent frequency.

[0090] This invention aims to specifically optimize the detection area on a hardware shaft caused by geometric abrupt changes such as keyways and steps. First, the unit, based on a digital twin model (CAD model) of the hardware shaft, calls a pre-built FEA solver. This solver accurately calculates the local resonant frequencies, i.e., natural frequencies, of these geometrically discontinuous regions under a specific excitation field, according to material properties.

[0091] Subsequently, when the encoder instructs the detection probe to enter the mismatched region, the unit initiates the genetic algorithm. The genetic algorithm uses the natural frequency calculated by the FEA as the target and performs global optimization within a preset frequency range. The fitness function of the algorithm is designed to maximize the matching degree between the excitation frequency and the natural frequency. After multiple generations of iterative selection, crossover, and mutation, the GA can quickly converge and output an optimal excitation frequency that maximizes the response in that region. The unit then instructs the excitation generator to switch to this frequency, thereby achieving adaptive, high-sensitivity detection of specific regions.

[0092] In some embodiments, it further includes: The intelligent diagnostic module is configured to: run a pre-trained autoencoder model, receive signal responses to calculate reconstruction errors, determine the initial excitation frequency based on the reconstruction errors, and extract the latent space feature vectors of the signal responses for online classification or identification of defects.

[0093] In this embodiment of the invention, an autoencoder model, such as a convolutional autoencoder, trained with a large amount of defect-free sample data is embedded. This model realizes the pattern of "healthy" signals.

[0094] During detection, the module receives real-time response signals and instructs the model to reconstruct them. By calculating the difference between the original signal and the reconstructed signal—the reconstruction error—the module can instantly determine the degree of signal abnormality. When the error exceeds a preset threshold, the module identifies the most significant distortion frequency caused by the defect and uses it as the optimal initial excitation frequency output, thereby achieving intelligent frequency selection.

[0095] Furthermore, for signals deemed anomalous, this module is also responsible for extracting their low-dimensional latent space feature vectors from the autoencoder output. This vector is a highly condensed "digital fingerprint" of the defect signal. This module can input these feature vectors into subsequent classifiers such as support vector machines, thereby achieving automatic online classification and identification of defect types, elevating detection from "detecting anomalies" to "diagnosing defects."

[0096] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. A method for quality inspection of precision hardware shafts, characterized in that, include: Based on the correspondence between the defect type and the excitation frequency of the hardware shaft, an initial excitation frequency for detecting the defect type is determined, and an excitation signal containing the initial excitation frequency is generated. The excitation signal is driven to act on the rotating hardware shaft, and the rotational speed and real-time angular position of the hardware shaft are obtained at the same time. Based on the rotational speed and the real-time angular position, the output rhythm, amplitude or phase of the excitation signal is synchronized and calibrated. The system monitors the signal response based on the synchronized and calibrated excitation signal in real time, and performs a first feedback adjustment on the frequency of the excitation signal when the characteristic value of the signal response deviates from the preset range, so as to enhance the detection capability of the signal response. The matching degree between the geometric position of the hardware shaft and the frequency of the excitation signal after the first feedback adjustment is analyzed, and when there is a mismatch area between the two, the frequency of the excitation signal is adjusted by the second feedback to adapt to the detection requirements of the mismatch area. Continuously collect defect response data based on the excitation signal adjusted by the second feedback, and adjust the frequency scanning step size or angle sampling density of the excitation signal by the third feedback according to whether there are missing frequency bands or angle ranges in the defect response data, so as to cover the missing frequency bands or angle ranges.

2. The method according to claim 1, characterized in that, The determination of the initial excitation frequency based on the correspondence between the defect type and the excitation frequency of the hardware shaft includes: From a database containing historical capacitance response data of different defect types at different excitation frequencies, the optimal excitation frequency range corresponding to the defect type to be tested is queried and determined, and the initial excitation frequency is selected from the range.

3. The method according to claim 1, characterized in that, The synchronization and calibration of the output rhythm, amplitude, or phase of the excitation signal includes: A phase-locked loop algorithm is used to lock the output frequency of the excitation signal to an integer multiple of the rotation frequency of the hardware shaft in order to achieve synchronization of the output rhythm; For each real-time angular position, the amplitude calibration and phase calibration values ​​for that angular position are calculated using linear interpolation and then applied to the excitation signal.

4. The method according to claim 1, characterized in that, The first feedback adjustment of the frequency of the excitation signal includes: When the characteristic value of the signal response deviates from the preset range, a gradient descent algorithm is used to iteratively adjust the frequency value of the current excitation signal with the goal of maximizing the deviation of the characteristic value.

5. The method according to claim 1, characterized in that, The second feedback adjustment of the frequency of the excitation signal includes: The natural frequency of the mismatched region is calculated using the finite element method, and then, combined with a genetic algorithm, the measured frequency is adjusted to an excitation frequency that is compatible with the natural frequency.

6. The method according to claim 1, characterized in that, After determining the initial excitation frequency, the method further includes: The real-time response signal of the tested hardware shaft is input into a pre-trained autoencoder model for reconstructing the response signal of the defect-free hardware shaft; Calculate the reconstruction error between the real-time response signal and the model reconstruction output signal; The frequency component that generates an error greater than a preset threshold is determined as the initial excitation frequency.

7. The method according to claim 6, characterized in that, The method further includes: Extract the latent space feature vector of the real-time response signal that generates an error greater than the preset error threshold in the autoencoder model.

8. The method according to claim 7, characterized in that, The method further includes: By performing unsupervised clustering on the collected latent space feature vectors, online classification of different defects can be achieved.

9. The method according to claim 8, characterized in that, The method further includes: Based on the manual labeling of the online classification results, the autoencoder model is updated using a transfer learning method, enabling the model to recognize the labeled defect types.

10. A quality inspection system for precision hardware shafts, characterized in that, include: The excitation and synchronization calibration unit is configured to: determine the initial excitation frequency according to the correspondence between the defect type of the hardware shaft and the excitation frequency; generate and drive an excitation signal containing the initial excitation frequency to act on the rotating hardware shaft; simultaneously acquire the rotation speed and real-time angular position of the hardware shaft; and synchronize and calibrate the output rhythm, amplitude or phase of the excitation signal according to the rotation speed and the real-time angular position. The first feedback control unit is connected to the excitation and synchronization calibration unit and is configured to: monitor the signal response based on the synchronized and calibrated excitation signal in real time, and when the characteristic value of the signal response deviates from the preset range, issue an instruction to the excitation and synchronization calibration unit to perform a first feedback adjustment on the frequency of the excitation signal; The second feedback control unit, connected to the first feedback control unit, is configured to: analyze the matching degree between the geometric position of the hardware shaft and the frequency of the excitation signal adjusted by the first feedback, and when there is a mismatch area between the two, issue a command to the excitation and synchronization calibration unit to perform a second feedback adjustment on the frequency of the excitation signal; The third feedback control unit, connected to the second feedback control unit, is configured to: continuously analyze the defect response data based on the excitation signal adjusted by the second feedback, and when there are frequency bands or angle ranges with missing signals in the defect response data, issue a command to the excitation and synchronization calibration unit to adjust the frequency scan step size or angle sampling density of the excitation signal.

Images