An unmanned intelligent production method for automobile fans

Abstract

The present invention discloses an unmanned intelligent production method for automotive fans. The method includes performing a low-speed spin-up process on rotating blades entering a constant-pressure flow channel, obtaining time-series data on the temperature rise of the axially distributed shear heat layer through segmented pressurization and three-stage speed increase, and establishing a baseline temperature rise curve. Differential processing is performed on the wall temperature rise data during the blade speed increase process to extract the temperature rise curvature exceeding the limit and generate abnormal annular zone coordinate data. The submillimeter wave oblique projection direction is determined based on the abnormal coordinates, and the echo signal is subjected to refraction phase analysis processing to obtain a set of warping fingerprint parameters including the refraction peak, echo tail, slope change rate, and phase oscillation period. Acoustic sampling is activated in the corresponding area to obtain air-acoustic resonance standing wave data, which is cross-validated with the warping fingerprint parameters to obtain quantitative detection and judgment results of blade geometric defects. This solution can achieve high-precision online detection of microscopic defects in fan blades in a completely dark factory environment.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned intelligent manufacturing and detection of automobile fans, and in particular to an unmanned intelligent production method for automobile fans. Background Art

[0002] Unmanned intelligent production of automotive fans utilizes automated equipment, intelligent sensor networks, and automated guided vehicle (AGV) logistics systems to automate the entire manufacturing process, from components to finished products. In this production model, the geometric accuracy and dynamic balancing quality of the blades directly determine the final aerodynamic performance. Even micron-level warpage or gram-level mass distribution deviations can induce abnormal vibration and noise during high-speed rotation, impacting the vehicle's NVH (Noise, Vibration, and Harshness) performance. Therefore, high-precision online inspection mechanisms must be implemented throughout the production process to promptly identify and remove potentially defective blades.

[0003] As smart manufacturing continues to evolve, the industry is moving toward a "dark factory": fully automated production, requiring no human intervention, with all visible light turned off, maintaining darkness in the workshop and relying solely on the equipment's built-in sensing and communication systems. This significantly reduces energy consumption and eliminates interference from ambient light on precision optical components. However, once in total darkness, traditional optical inspection methods such as laser scanning and machine vision become ineffective. Real-time, online detection of microscopic defects in high-speed rotating blades in a completely dark environment has become a technical pain point that urgently needs to be overcome in this field. Summary of the Invention

[0004] The main purpose of the present invention is to solve the technical problem that the existing technology cannot realize real-time online detection of microscopic defects of high-speed rotating blades in a dark environment.

[0005] A first aspect of the present invention provides an unmanned intelligent production method for automobile fans, the unmanned intelligent production method for automobile fans comprising:

[0006] Perform low-speed rotation on the rotating blades entering the constant-pressure flow channel, continuously obtain the time series data of the shear heat layer temperature rise on the flow channel wall, and obtain the benchmark temperature rise curve;

[0007] According to the reference temperature rise curve, differential processing is performed on the time series data of the wall shear heat layer temperature rise acquired in real time during the blade acceleration process, the temperature rise curvature exceeding limit area is extracted and the abnormal ring coordinate data is generated;

[0008] Determine the oblique projection direction of the submillimeter wave according to the coordinate data of the abnormal ring zone, perform refraction phase analysis on the echo signal after the submillimeter wave passes through the abnormal ring zone, and obtain a warping fingerprint parameter set;

[0009] According to the abnormal annular coordinate data, acoustic sampling of the corresponding area is activated to obtain the air-acoustic resonance standing wave data generated by the blade rotation. The air-acoustic resonance standing wave data is cross-validated with the warping fingerprint parameter set to obtain the quantitative detection and judgment results of the blade geometric defects.

[0010] Preferably, the low-speed rotation start process is performed on the rotating blades entering the constant-pressure flow channel, and the time series data of the shear heat layer temperature rise of the flow channel wall is continuously obtained to obtain the reference temperature rise curve, including:

[0011] Perform segmented pressurization on the gas in the inlet, middle, and outlet sections of the flow channel to obtain axial pressure gradient distribution data;

[0012] According to the axial pressure gradient distribution data, a three-stage rotation speed increasing process is performed on the rotating blade to obtain inlet section shear response data, middle section shear response data, and outlet section shear response data;

[0013] Perform real-time temperature rise sampling processing on the inlet section wall, middle section wall, and outlet section wall to obtain the inlet section temperature rise time series data, middle section temperature rise time series data, and outlet section temperature rise time series data;

[0014] performing weighted fusion processing on the inlet section temperature rise time series data, the middle section temperature rise time series data, and the outlet section temperature rise time series data according to the axial pressure gradient distribution data, the inlet section shear response data, the middle section shear response data, and the outlet section shear response data to obtain axial comprehensive temperature rise time series data;

[0015] An attenuation tail feature extraction process is performed on the axial comprehensive temperature rise time series data to obtain a reference temperature rise curve including a temperature rise peak value, an attenuation constant, and a tail period.

[0016] Preferably, the three-stage rotation speed increasing process is performed on the rotating blade according to the axial pressure gradient distribution data to obtain the inlet section shear response data, the middle section shear response data, and the outlet section shear response data, including:

[0017] Perform acceleration curve planning processing based on the blade radial moment of inertia distribution data and the blade radial natural vibration frequency data to obtain three-segment angular acceleration sequence data;

[0018] Performing a first-stage speed increasing process according to the three-segment angular acceleration sequence data, acquiring inlet section shear response time series data when the blade speed reaches the inlet section target speed, and obtaining inlet section shear response data;

[0019] Performing a second stage of speed increasing processing according to the three-segment angular acceleration sequence data, acquiring intermediate segment shear response time series data when the blade speed reaches the intermediate segment target speed, and obtaining intermediate segment shear response data;

[0020] The third stage speed increasing processing is performed according to the three sections of angular acceleration sequence data, and when the blade speed reaches the outlet section target speed, the outlet section shear response time series data is acquired to obtain the outlet section shear response data.

[0021] Preferably, performing weighted fusion processing on the inlet section temperature rise time series data, the middle section temperature rise time series data, and the outlet section temperature rise time series data according to the axial pressure gradient distribution data and the inlet section shear response data, the middle section shear response data, and the outlet section shear response data to obtain the axial comprehensive temperature rise time series data includes:

[0022] Normalizing the blade radial mass distribution data and the blade moment of inertia parameters, and combining them with the axial pressure gradient distribution data to obtain a segment weight basis matrix;

[0023] performing response intensity weighting processing on the inlet segment shear response data, the middle segment shear response data, and the outlet segment shear response data according to the segment weight basic matrix to obtain a segment dynamic weight coefficient vector;

[0024] According to the segment dynamic weight coefficient vector, multi-scale wavelet weighted fusion processing is performed on the inlet segment temperature rise time series data, the middle segment temperature rise time series data, and the outlet segment temperature rise time series data to obtain axial comprehensive temperature rise time series data.

[0025] Preferably, the step of performing differential processing on the wall shear heat layer temperature rise time series data acquired in real time during the blade acceleration process based on the reference temperature rise curve, extracting the temperature rise curvature exceeding limit area and generating abnormal annular zone coordinate data includes:

[0026] Perform time base synchronization processing on the real-time acquired wall shear heat layer temperature rise time series data and blade speed time series data to obtain temperature rise and speed synchronization sequence data;

[0027] Performing first-order difference processing on the temperature rise and speed synchronization sequence data according to the reference temperature rise curve to obtain temperature rise deviation sequence data;

[0028] Performing second-order differential curvature calculation processing on the temperature rise deviation sequence data to obtain temperature rise curvature time series data;

[0029] Speed-related threshold judgment processing is performed on the temperature-rise curvature time series data and the blade speed time series data to obtain temperature-rise curvature excess area data, and coordinate mapping processing is performed on the temperature-rise curvature excess area data according to the axial coordinate data of the flow channel wall to obtain abnormal annular zone coordinate data.

[0030] Preferably, the speed-related threshold determination processing is performed on the temperature-rise curvature time series data and the blade speed time series data to obtain the temperature-rise curvature limit-exceeding region data; and the coordinate mapping processing is performed on the temperature-rise curvature limit-exceeding region data according to the axial coordinate data of the flow channel wall to obtain the abnormal annular zone coordinate data, including:

[0031] Performing time integration processing on the blade rotation speed time series data to obtain blade angular position time series data;

[0032] Performing speed-related threshold determination processing based on the temperature-rise curvature time series data and the blade angle position time series data to obtain temperature-rise curvature exceeding limit region data;

[0033] Performing polar coordinate mapping processing on the temperature rise curvature exceeding limit area data and the blade angular position time series data to obtain polar coordinate sequence data of the blade outer surface;

[0034] Circumferential stitching processing is performed on the polar coordinate sequence data of the blade outer surface according to the blade circumferential pitch data, and axial mapping processing is performed on the axial coordinate data of the flow channel wall to obtain abnormal annular zone coordinate data.

[0035] Preferably, the submillimeter wave oblique projection direction is determined according to the abnormal annular zone coordinate data, and a refraction phase analysis process is performed on the echo signal after the submillimeter wave passes through the abnormal annular zone to obtain a warping fingerprint parameter set, including:

[0036] Performing oblique projection angle band calculation processing based on the abnormal annular zone coordinate data and the blade radial dielectric constant distribution data to obtain submillimeter wave oblique projection direction data;

[0037] Performing oblique transmission control processing on the submillimeter wave beam according to the submillimeter wave oblique projection direction data to obtain submillimeter wave echo original time series data;

[0038] performing time normalization processing on the submillimeter wave echo original time series data according to the blade angle position time series data to obtain submillimeter wave echo normalized sequence data;

[0039] performing refraction phase unwrapping processing on the submillimeter wave echo normalized sequence data to obtain refraction phase sequence data;

[0040] Peak extraction, tail attenuation estimation, phase slope calculation and oscillation period determination processing are performed on the refraction phase sequence data to obtain a warping fingerprint parameter set.

[0041] Preferably, performing oblique projection angle band calculation processing based on the abnormal annular zone coordinate data and the blade radial dielectric constant distribution data to obtain submillimeter wave oblique projection direction data includes:

[0042] Performing coordinate matching processing on blade radial thickness distribution data according to the abnormal annular zone coordinate data to obtain annular zone thickness sequence data;

[0043] Performing attenuation compensation calculation processing based on the annular belt thickness sequence data and the blade radial dielectric constant distribution data to obtain annular belt effective penetration depth data;

[0044] Performing bandwidth screening processing on the available submillimeter wave spectrum interval according to the annular band effective penetration depth data to obtain annular band candidate frequency band data;

[0045] Angle and frequency band joint optimization processing is performed according to the candidate frequency band data of the ring band and the abnormal coordinate data of the abnormal ring band to obtain submillimeter wave oblique projection direction data.

[0046] Preferably, activating acoustic sampling of a corresponding area according to the abnormal annular zone coordinate data, obtaining air-acoustic resonance standing wave data generated by blade rotation, cross-validating the air-acoustic resonance standing wave data with the warping fingerprint parameter set, and obtaining a quantitative detection and determination result of blade geometric defects includes:

[0047] Determining the acoustic sampling window coordinate data according to the abnormal ring coordinate data;

[0048] Acquiring air-acoustic resonance standing wave time series data according to the acoustic sampling window coordinate data to obtain air-acoustic resonance standing wave data;

[0049] performing rotation order demodulation processing on the air-acoustic resonance standing wave data to obtain order energy spectrum data;

[0050] Performing amplitude-phase mapping processing on the order energy spectrum data according to the blade pitch angle distribution data to obtain pitch angle mapping spectrum data;

[0051] A feature coupling judgment process is performed based on the pitch angle mapping spectrum data and the warping fingerprint parameter set to obtain a quantitative detection and judgment result of the blade geometric defect.

[0052] A second aspect of the present invention provides an unmanned intelligent production line for automobile fans, which adopts the unmanned intelligent production method for automobile fans described in any of the above embodiments.

[0053] The technical solution provided in the embodiment of the present application continuously records the temperature rise-time curve on the shear heat layer naturally formed on the wall when the blade just enters the flow channel and the rotational speed is still low, and locks this curve as the aerodynamic-thermal baseline of the current batch with the help of constant pressure and dark field conditions. As the blade accelerates along the set speed-up curve, the wall temperature rise sequence is synchronously updated, and the system subtracts the real-time sequence from the baseline. Any excess of the second-order curvature is immediately mapped to the composite coordinates of the flow channel axis and the blade angular position, thereby outlining the abnormal annulus. The speed-up stage takes only a few hundred milliseconds, and the differential operation occurs directly at the physical quantity level, avoiding large-scale optical scanning of the entire blade surface and compressing the detection window from the source.

[0054] Once the anomaly annulus is identified, the submillimeter wave beam traverses it tangentially at a calculated tilt angle and frequency band. The beam generates a phase delay within the shear heat layer and the blade air gap, which is monotonically dependent on the warping amplitude and local mass offset. After phase unwrapping, the real-time echo outputs four dimensionless parameters, forming a warping fingerprint that contains both geometric information and implicit moment of inertia changes. This fingerprint generation process leverages the high sensitivity of submillimeter waves to microscale refractive index gradients, enabling the resolution of micron-level deformations without the aid of visible light.

[0055] The space where the same abnormal annulus is located is the acoustic sampling window. The shear layer airflow induced by the blade rotation couples with the flow channel resonance cavity to generate standing waves with clear orders. After the acoustic array obtains the energy spectrum, it is mathematically coupled with the phase slope and oscillation period in the warping fingerprint. Only when the two are consistent in both amplitude and phase dimensions does the system give a judgment that the defect is established. Thermal field differentials provide position constraints, submillimeter waves provide deformation variables, and air-acoustic resonance supplements the dynamic response verification. The three are mutually checked. This circumvents the signal-to-noise bottleneck of traditional visual or vibration methods in a dark environment, realizes online quantitative judgment of micro-warping and quality unevenness, and reduces the false alarm rate to an acceptable range for the process through a crossover mechanism, thereby maintaining a stable dynamic balance yield on the all-black unmanned production line. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the structures shown in these drawings without paying any creative work.

[0057] Figure 1 This is a schematic diagram of an embodiment of the unmanned intelligent production method for automobile fans in an embodiment of the present invention.

[0058] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0059] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0060] It should be noted that if the embodiments of the present invention involve directional indications (such as up, down, left, right, front, back, etc.), the directional indications are only used to explain the relative position relationship, movement status, etc. between the various components under a certain specific posture. If the specific posture changes, the directional indications will also change accordingly.

[0061] In addition, the descriptions of "first", "second", etc. in the present invention are only for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In addition, "and / or" in the full text includes three solutions. Taking A and / or B as an example, it includes technical solution A, technical solution B, and technical solution that satisfies both A and B. In addition, the technical solutions between the various embodiments can be combined with each other, and must be based on the ability of ordinary technicians in this field to implement. When the combination of technical solutions is mutually contradictory or cannot be implemented, it should be deemed that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0062] An embodiment of the present application provides an unmanned intelligent production method for automobile fans. Figure 1 A flowchart of an unmanned intelligent production method for automobile fans provided in one embodiment of the present application. In this embodiment, the method includes:

[0063] See also Figure 1 , perform low-speed rotation processing on the rotating blades entering the constant-pressure flow channel, continuously obtain the time series data of the shear heat layer temperature rise on the flow channel wall, and obtain the benchmark temperature rise curve;

[0064] In one embodiment of the present invention, performing low-speed rotation start processing on the rotating blades entering the constant-pressure flow channel, continuously acquiring time series data of the shear heat layer temperature rise on the flow channel wall, and obtaining a reference temperature rise curve includes:

[0065] Perform segmented pressurization on the gas in the inlet, middle, and outlet sections of the flow channel to obtain axial pressure gradient distribution data;

[0066] According to the axial pressure gradient distribution data, a three-stage rotation speed increasing process is performed on the rotating blade to obtain inlet section shear response data, middle section shear response data, and outlet section shear response data;

[0067] Perform real-time temperature rise sampling processing on the inlet section wall, middle section wall, and outlet section wall to obtain the inlet section temperature rise time series data, middle section temperature rise time series data, and outlet section temperature rise time series data;

[0068] performing weighted fusion processing on the inlet section temperature rise time series data, the middle section temperature rise time series data, and the outlet section temperature rise time series data according to the axial pressure gradient distribution data, the inlet section shear response data, the middle section shear response data, and the outlet section shear response data to obtain axial comprehensive temperature rise time series data;

[0069] An attenuation tail feature extraction process is performed on the axial comprehensive temperature rise time series data to obtain a reference temperature rise curve including a temperature rise peak value, an attenuation constant, and a tail period.

[0070] The following is a detailed description of the steps involved in the above embodiment:

[0071] As an optional embodiment, the constant-pressure channel can be a circular channel with an axial length of 3 meters and an inner diameter of 0.8 meters, designed to accommodate rotating blades for testing. The channel is divided axially into three equal sections: the inlet section, which represents the first 1-meter area immediately after the blades enter the channel; the middle section, which represents the middle 1-meter area where the blades rotate steadily; and the outlet section, which represents the last 1-meter area immediately before the blades exit the channel. Each section is connected to the adjacent section by an adjustable baffle, forming a relatively independent pressure chamber. The baffle has a circular hole in the center that matches the outer diameter of the blades, allowing free passage of the blades but restricting significant gas flow. Each section is equipped with an independent gas supply system and pressure regulator, with pressure controlled by precision pressure regulating valves. The inlet section is connected to a high-pressure gas source and is set to 1.2 standard atmospheres; the middle section is connected to a standard-pressure gas source and is set to 1.0 standard atmospheres; and the outlet section is connected to a low-pressure gas source and is set to 0.8 standard atmospheres. The flow-restricting apertures in the baffles ensure that only trace amounts of gas exchange between the sections are present, insufficient to eliminate pressure gradients. Pressure sensors monitor the pressure values ​​at each section in real time. When the pressure deviation exceeds ±0.02 atmospheres, an automatic control device immediately replenishes or vents gas to maintain the set pressure. The axial pressure gradient distribution data refers to the sequence of pressure changes along the axial direction from the inlet to the outlet. Pressure values ​​are recorded at sampling intervals of 10 centimeters, forming a 30-point pressure distribution array. This segmented pressure control method creates testing conditions where the blade faces different pressure environments at different axial positions, resulting in differentiated characteristics of the shear heat layer in each section.

[0072] The system determines the phased speed strategy based on the pressure environment of each section and the aerodynamic characteristics of the blade. The blade radial moment of inertia distribution data refers to the numerical distribution of the moment of inertia of the blade at each radial position from the root to the top, which is obtained through blade mass measurement and geometric scanning. The blade radial natural vibration frequency data refers to the natural vibration frequency distribution of the blade at each radial position, which is obtained through modal analysis testing. The three-segment angular acceleration sequence data is the optimal angular acceleration value of each stage calculated based on the inertia distribution and natural frequency to avoid exciting blade resonance. The servo motor controller drives the blade to rotate according to the calculated angular acceleration curve: the first stage is at 30 rpm 2 The angular acceleration is accelerated from rest to 600 rpm in 20 seconds; the second stage is accelerated at 25 rpm. 2 The angular acceleration is accelerated from 600 rpm to 900 rpm in 12 seconds; the third stage is accelerated at 20 rpm. 2 The angular acceleration is accelerated from 900 rpm to 1200 rpm in 15 seconds. When the blade speed stabilizes at the target speed of each section, the pressure pulsation sensor arranged on the wall of the corresponding flow channel starts to record data. Shear response data refers to the pressure pulsation signal caused by the shear effect of the airflow generated by the rotation of the blade on the wall of the flow channel, which reflects the aerodynamic distribution characteristics of the blade surface. Each data acquisition lasts for 5 seconds, with a sampling frequency of 2000Hz, and shear response time series data of 10,000 data points is obtained. The staged speed increase avoids the impact disturbance of the flow field caused by the instantaneous acceleration of the blade, ensuring the stability and independence of the shear response data of each section.

[0073] Twenty-four thermocouple temperature sensors are arranged in an 8×3 matrix along the inner channel wall in each axial section, with each sensor spaced 12.5 cm apart. These sensors use contact thermocouple temperature measurement, measuring wall temperature through direct contact between the sensor probe and the channel wall. The measurement accuracy is ±0.005°C, and the response time is less than 1 millisecond. The temperature data acquisition system is synchronized with the blade speed encoder signal to ensure accurate temporal correspondence between temperature rise data and speed changes. The temperature rise time series data for each section includes ambient temperature baseline subtraction and digital filtering. First, the ambient temperature value at each sampling point when the blade is stationary is subtracted as a zero-point correction. Then, a Butterworth low-pass filter is used to remove high-frequency noise above 500 Hz. The inlet, middle, and outlet section temperature rise time series data are each a matrix of temperature rise values ​​over time at 24 sampling points in each section. Each matrix contains three-dimensional information: time, spatial position, and temperature rise value. The completely dark environment eliminates the interference of external infrared radiation, allowing the weak temperature rise signal (usually only 0.3-0.8℃) generated by the shear heat layer to be accurately detected, and the signal-to-noise ratio can be effectively improved compared to the light environment.

[0074] The segment weight base matrix is ​​generated using a multi-parameter normalization algorithm. The radial mass distribution data of the blade is obtained by segmented weighing using a precision balance. The blade is divided into 10 equal-length segments along the radial direction, and the mass of each segment is measured separately. The blade moment of inertia parameter is measured using a three-wire pendulum measuring device to obtain the overall moment of inertia value. Normalization is performed by dividing the mass distribution data and the moment of inertia parameter by their maximum values, resulting in dimensionless values ​​in the range of 0-1. The axial pressure gradient distribution data is also normalized, with the pressure values ​​of each segment divided by the inlet segment pressure value. The segment weight base matrix is ​​composed of a 3×3 matrix consisting of three types of data: normalized mass weight, inertia weight, and pressure weight. The segment dynamic weight coefficient vector is obtained through response intensity weighting: the root mean square value of the shear response data of each segment is calculated as the response intensity index, and the response intensity is multiplied by the segment weight base matrix to obtain the dynamic weight coefficient that takes into account the physical characteristics and real-time response state of the blade. Multiscale wavelet weighted fusion processing uses Daubechies4 wavelets to perform a three-layer decomposition of the temperature rise data for each segment. Dynamic weight coefficients are applied to each scale, and then reconstructed to obtain the axial integrated temperature rise time series data. This weighted fusion method fully considers the impact of uneven blade mass distribution and the differences in the aerodynamic environment between segments on the temperature rise pattern, avoiding the problem that simple arithmetic averaging can obscure important information.

[0075] The attenuation tail feature extraction is achieved through time domain signal analysis. The temperature rise peak is obtained by searching for the maximum value point in the axial integrated temperature rise time series data through the global maximum search algorithm. The attenuation constant calculation selects the data segment that continuously decreases after the peak, and adopts nonlinear least squares fitting of the exponential decay function, and the fitting window length is the data 2 seconds after the peak. The tail period is determined by power spectrum density analysis: the attenuation segment data is fast Fourier transformed to find the dominant frequency component in the power spectrum, and its reciprocal is the tail period. The benchmark temperature rise curve is fully described by three characteristic parameters: temperature rise peak, attenuation constant, and tail period, forming the thermodynamic benchmark fingerprint of the current blade batch. This parameterized characterization method compresses the complex temperature rise time series signal into three key physical quantities, which not only retains the main dynamic characteristics of the shear heat layer, but also provides a standardized comparison benchmark for subsequent anomaly detection, ensuring the high accuracy and high efficiency of the detection process in a completely dark environment.

[0076] In one embodiment of the present invention, performing a three-stage rotation speed increasing process on the rotating blade according to the axial pressure gradient distribution data to obtain inlet section shear response data, middle section shear response data, and outlet section shear response data includes:

[0077] Perform acceleration curve planning processing based on the blade radial moment of inertia distribution data and the blade radial natural vibration frequency data to obtain three-segment angular acceleration sequence data;

[0078] Performing a first-stage speed increasing process according to the three-segment angular acceleration sequence data, acquiring inlet section shear response time series data when the blade speed reaches the inlet section target speed, and obtaining inlet section shear response data;

[0079] Performing a second stage of speed increasing processing according to the three-segment angular acceleration sequence data, acquiring intermediate segment shear response time series data when the blade speed reaches the intermediate segment target speed, and obtaining intermediate segment shear response data;

[0080] The third stage speed increasing processing is performed according to the three sections of angular acceleration sequence data, and when the blade speed reaches the outlet section target speed, the outlet section shear response time series data is acquired to obtain the outlet section shear response data.

[0081] The following is a detailed description of the steps involved in the above embodiment:

[0082] The specific implementation process of the acceleration curve planning process is as follows: First, the blade geometry is scanned by a three-dimensional coordinate measuring machine, and the blade is divided into 20 equal sections along the radial direction. The thickness, chord length, and torsion angle parameters of each section are measured. Combined with the blade material density of 2.7g / cm 3 Calculate the moment of inertia value of each section to form the radial moment of inertia distribution data of the blade. Then use a modal tester to measure the radial natural vibration frequency data of the blade: fix the blade on the fixture, and use a hammer to hit each radial position in turn. The acceleration sensor collects the vibration response signal, and the spectrum analyzer performs a fast Fourier transform to obtain the first-order bending natural frequency of each position. Next, the angular acceleration is calculated: the maximum allowable angular acceleration of each stage is determined based on the moment of inertia distribution, and the safe angular acceleration range to avoid resonance is determined based on the natural frequency distribution, and the minimum value of the two is taken as the actual angular acceleration. The three-segment angular acceleration sequence data is obtained through numerical calculation: the first stage is 25 rpm 2 , second stage 20 rpm 2 , the third stage 15 rpm 2 For example, the moment of inertia of the blade root is 0.08 kg·m 2 , When the natural frequency is 45Hz, the system calculates the maximum allowable angular acceleration to be 30 rpm 2 , the safety angular acceleration is 25 rpm 2 , take the smaller value 25 rpm 2 This decreasing angular acceleration design avoids structural stress concentration caused by excessive acceleration at high blade speeds, while also avoiding the natural vibration frequencies at each radial position, ensuring the stability of the shear heat layer formed during the acceleration process.

[0083] The first stage of speed increase is realized by using a frequency converter to drive the servo motor. The frequency converter is set to 25 rpm. 2The angular acceleration outputs a corresponding frequency signal, which the servo motor uses to rotate the blades from a stationary position. A speed encoder mounted on the motor shaft monitors the blade speed in real time with a resolution of 1000 pulses / rev. The encoder signal is converted by a counter into a speed value and fed back to the inverter control system. When the speed reaches the target inlet speed of 600 rpm and remains stable within ±5 rpm for 2 seconds, the programmable logic controller triggers data acquisition. Time series data on the inlet shear response are collected using eight differential pressure sensors evenly distributed along the inlet channel wall. These sensors measure the pressure difference between the channel wall and the ambient atmosphere. The data acquisition card continuously collects data at 2000 Hz for 5 seconds, acquiring 10,000 pressure data points. The raw data is first zero-corrected by subtracting the sensor output value when the blades are stationary. The data is then band-pass filtered at 10-500 Hz by the digital signal processor to remove DC drift and high-frequency electromagnetic interference. The inlet section shear response data consists of three statistical parameters: the RMS value represents the average intensity of the pressure pulsation, the crest factor indicates the sharpness of the pulsation, and the skewness indicates the symmetry of the pressure distribution. This precise speed control and multi-parameter data extraction method fully characterizes the shear characteristics of the blade in the inlet section aerodynamic environment, providing reliable foundational data for subsequent multi-segment data fusion.

[0084] The second and third stages of the speed increase process use the same control method, with only the target speed and sensor parameter settings being different. In the second stage, the angular acceleration is adjusted to 20 rpm. 2 The target speed is set to 900 rpm, and the differential pressure sensor range is adjusted to ±2kPa to adapt to the increased pressure pulsation amplitude. The third stage angular acceleration is reduced to 15 rpm 2 , the target speed is increased to 1200 rpm, the sensor range is expanded to ±3kPa, and the sampling frequency is increased to 3000Hz to capture higher-frequency pressure pulsation components at high speeds. The statistical parameter calculation method for the intermediate section shear response data and the outlet section shear response data is exactly the same as that in the first stage, ensuring the comparability of the three sections of data. The technical effect of the decreasing angular acceleration strategy is to adapt to the changes in the dynamic characteristics of the blade at different speeds: at low speeds, the blade is more rigid and can withstand larger angular accelerations. At high speeds, the centrifugal force increases and the angular acceleration needs to be reduced to avoid overstress. At the same time, the three target speeds of 600, 900, and 1200 rpm correspond to the three typical operating conditions of low, medium, and high speed blades, respectively, covering the main speed range of the blade in actual use, and providing a complete data basis for a comprehensive evaluation of the shear characteristics of the blade.

[0085] In one embodiment of the present invention, the weighted fusion processing is performed on the inlet section temperature rise time series data, the middle section temperature rise time series data, and the outlet section temperature rise time series data according to the axial pressure gradient distribution data and the inlet section shear response data, the middle section shear response data, and the outlet section shear response data to obtain the axial comprehensive temperature rise time series data, including:

[0086] Normalizing the blade radial mass distribution data and the blade moment of inertia parameters, and combining them with the axial pressure gradient distribution data to obtain a segment weight basis matrix;

[0087] performing response intensity weighting processing on the inlet segment shear response data, the middle segment shear response data, and the outlet segment shear response data according to the segment weight basic matrix to obtain a segment dynamic weight coefficient vector;

[0088] According to the segment dynamic weight coefficient vector, multi-scale wavelet weighted fusion processing is performed on the inlet segment temperature rise time series data, the middle segment temperature rise time series data, and the outlet segment temperature rise time series data to obtain axial comprehensive temperature rise time series data.

[0089] The following is a detailed description of the steps involved in the above embodiment:

[0090] The generation of the segment weight basic matrix is ​​achieved through normalized numerical calculation and matrix construction algorithm. The process of obtaining the radial mass distribution data of the blade is as follows: the blade is divided into 10 segments of equal length along the radial direction. The mass of each segment can be obtained through production-related data, or the mass of each segment can be measured separately using a precision electronic balance to obtain the mass distribution array from the blade root to the blade tip in grams. The blade moment of inertia parameter is measured by a three-wire pendulum test device to measure the moment of inertia of the entire blade in kg·m 2. The specific calculation process of the normalization processing is: divide each value in the mass distribution data by the maximum mass value, and divide the moment of inertia parameter by the product of the total mass of the blade and the square of the blade radius to obtain a dimensionless value in the range of 0-1. The axial pressure gradient distribution data is also normalized: the pressure values ​​of the inlet section, the middle section, and the outlet section are divided by the inlet section pressure value to obtain the relative pressure ratio. The segment weight basic matrix is ​​a 3×3 numerical matrix, the rows represent the three axial sections, and the columns represent the three types of parameters: mass weight, inertia weight, and pressure weight. The specific values ​​of the matrix elements are calculated by weighted average: the mass weight of each row is the average value of the normalized mass of each radial position in the corresponding section, the inertia weight is the normalized moment of inertia parameter, and the pressure weight is the normalized pressure value of the corresponding section. For example, the inlet section contains three radial positions at the root of the blade, and the corresponding normalized masses are 0.8, 0.9, and 1.0, respectively. The mass weight of this section is 0.9. This multi-parameter normalized matrix fully considers the combined effects of uneven blade mass distribution and differences in pressure environment in each section on the temperature rise pattern, avoids the limitations of single parameter weight allocation, and provides a comprehensive physical basis for subsequent dynamic weight calculation.

[0091] The segment dynamic weight coefficient vector is obtained through response intensity quantification and matrix operations. The specific implementation steps for response intensity weighting are as follows: First, the response intensity index of each segment shear response data is calculated. Specifically, the root mean square (RMS) value of the inlet, middle, and outlet shear response data is calculated to reflect the average energy level of the shear response in each segment. The three RMS values ​​are then used to form a response intensity vector, which is normalized by dividing each element by the maximum value of the vector to obtain a response intensity weight within the range of 0–1. Next, a matrix multiplication operation is performed: the 3×3 segment weight basis matrix is ​​multiplied by the 3×1 response intensity weight vector to obtain a 3×1 segment dynamic weight coefficient vector. The segment dynamic weight coefficient vector contains three elements, corresponding to the dynamic weight coefficients of the inlet, middle, and outlet segments, respectively, and has a value range of 0–1. The calculation process uses the numerical computing software MATLAB to perform matrix operations to ensure accuracy and consistency. For example, when the inlet segment response strength is 0.8, the middle segment is 1.0, and the outlet segment is 0.6, the response strength vector is [0.8, 1.0, 0.6]. Multiplying this with the segment weight matrix yields the segment dynamic weight coefficient vector [0.75, 0.90, 0.65]. This dynamic weight adjustment mechanism, based on real-time response strength, adaptively adjusts the contribution of each segment data based on the blade's current shear characteristics, more accurately reflecting the blade's actual physical state than fixed weight assignments.

[0092] Multi-scale wavelet weighted fusion processing is implemented using discrete wavelet transforms and a weighted reconstruction algorithm. This digital signal processing method involves performing wavelet decomposition on a signal at multiple scales, applying weighting coefficients to each scale, and then reconstructing the fused signal. The specific implementation process is as follows: First, a three-layer discrete wavelet transform is performed on the inlet, middle, and outlet temperature rise time series data, using the Daubechies4 wavelet basis function to obtain wavelet coefficients at different frequency scales for each segment. Wavelet decomposition decomposes the original signal into low-frequency approximation coefficients and high-frequency detail coefficients. The three-layer decomposition yields one approximation coefficient group and three detail coefficient groups. Then, a segment dynamic weight coefficient vector is applied at each scale: the wavelet coefficients for each segment at the corresponding scale are multiplied by the corresponding dynamic weight coefficient. For example, the first-layer detail coefficients for the inlet segment are multiplied by 0.75, the middle segment by 0.90, and the outlet segment by 0.65. Next, the weighted wavelet coefficients for each segment are summed and fused to obtain the fused wavelet coefficients. Finally, the fused signal is reconstructed through the inverse discrete wavelet transform to obtain the axial comprehensive temperature rise time series data. The axial comprehensive temperature rise time series data is a single time series containing the combined contributions of three segments of temperature rise information, retaining the main characteristics of each segment while suppressing local noise. The three-layer setting of wavelet decomposition can effectively separate the temperature rise changes of different frequency components. The low-frequency components reflect the overall trend, and the high-frequency components reflect local fluctuations. The layered weighting ensures the accurate fusion of signal components with different physical meanings. Compared with simple weighted averaging, this multi-scale fusion method can better maintain the time-frequency characteristics of the temperature rise signal, avoid the loss of signal details, and provide high-quality comprehensive data for subsequent baseline curve feature extraction.

[0093] Please continue reading Figure 1 , according to the reference temperature rise curve, differential processing is performed on the time series data of the wall shear heat layer temperature rise acquired in real time during the blade acceleration process, the temperature rise curvature exceeding limit area is extracted and the abnormal ring coordinate data is generated;

[0094] In one embodiment of the present invention, the differential processing of the wall shear heat layer temperature rise time series data acquired in real time during the blade acceleration process based on the reference temperature rise curve, extracting the temperature rise curvature exceeding limit area and generating abnormal annular zone coordinate data includes:

[0095] Perform time base synchronization processing on the real-time acquired wall shear heat layer temperature rise time series data and blade speed time series data to obtain temperature rise and speed synchronization sequence data;

[0096] Performing first-order difference processing on the temperature rise and speed synchronization sequence data according to the reference temperature rise curve to obtain temperature rise deviation sequence data;

[0097] Performing second-order differential curvature calculation processing on the temperature rise deviation sequence data to obtain temperature rise curvature time series data;

[0098] Speed-related threshold judgment processing is performed on the temperature-rise curvature time series data and the blade speed time series data to obtain temperature-rise curvature excess area data, and coordinate mapping processing is performed on the temperature-rise curvature excess area data according to the axial coordinate data of the flow channel wall to obtain abnormal annular zone coordinate data.

[0099] The following is a detailed description of the steps involved in the above embodiment:

[0100] Time base synchronization is achieved using a digital signal synchronization algorithm and interpolation resampling techniques. The wall shear layer temperature rise time series data is collected at a frequency of 2000 Hz by thermocouple temperature sensors placed on the flow channel wall, while the blade speed time series data is collected at a frequency of 5000 Hz by a photoelectric encoder mounted on the motor shaft. These two data streams have different sampling frequencies and time starting points. Time base synchronization is a data preprocessing method that unifies multiple data streams with different sampling frequencies to the same time base and sampling frequency. The specific implementation process is as follows: First, a GPS timing module provides a unified timestamp reference for the two data acquisition systems, ensuring absolute time synchronization of the data recording. A lower sampling frequency of 2000 Hz is then selected as the unified sampling frequency. The blade speed time series data is downsampled: an anti-aliasing filter is used to remove high-frequency components, and then the data is resampled to a frequency of 2000 Hz. Next, time alignment is performed: based on the timestamp information of the two data streams, the temperature rise data and the speed data are precisely aligned on the time axis to eliminate the difference in acquisition start time. Finally, a linear interpolation algorithm is used to fill in the data gaps caused by the sampling frequency difference. The synchronized temperature and speed data sequences are two data sequences with the same sampling frequency and time base, obtained through time synchronization. These sequences contain both the temperature rise and speed values ​​at the corresponding instants. For example, at 0.001 seconds, the synchronized data sequences simultaneously record a temperature rise of 25.3°C and a speed of 587 rpm. This precise synchronization eliminates time deviations between different sensor acquisition systems, ensuring a strict correspondence between temperature and speed changes in subsequent differential analysis, and preventing misjudgments due to time misalignment.

[0101] First-order difference processing is achieved through a numerical differentiation algorithm and benchmark comparison calculations. The benchmark temperature rise curve contains three characteristic parameters: peak temperature rise, decay constant, and tailing period. These parameters are used to reconstruct a complete benchmark temperature rise time function. First-order difference processing is a data processing method that calculates the numerical difference between the real-time temperature rise data and the benchmark temperature rise function at corresponding moments. The specific calculation process is as follows: First, the corresponding speed value is extracted from the temperature rise and speed synchronization sequence data based on the current blade speed. Then, the theoretical temperature rise value at that speed is calculated based on the characteristic parameters of the benchmark temperature rise curve. Next, the difference between the measured temperature rise value and the theoretical temperature rise value is calculated to obtain the temperature rise deviation value at that moment. This calculation process is performed point by point at each moment in the synchronization sequence data, forming a complete temperature rise deviation time series. The temperature rise deviation series data is a data series consisting of the point-by-point difference between the measured temperature rise value and the benchmark temperature rise value. The values ​​are expressed in degrees Celsius. Positive values ​​indicate that the measured temperature rise exceeds the benchmark value, while negative values ​​indicate that the measured temperature rise is below the benchmark value. Data processing uses a DSP digital signal processor for real-time calculations, with a computational latency of less than 5 milliseconds. For example, if the baseline temperature rise at a certain moment is 25.0°C and the measured temperature rise is 25.4°C, the temperature rise deviation at that moment is +0.4°C. This speed-based dynamic benchmark comparison method more accurately identifies temperature rise anomalies that do not match the blade speed state than fixed threshold judgment, effectively eliminating temperature rise fluctuations caused by normal speed changes.

[0102] The second-order differential curvature calculation process uses a numerical second-order derivative algorithm to achieve quantitative analysis of the temperature rise change rate. The second-order differential curvature calculation process refers to performing a second-order differential operation on the first-order differential result to obtain a second-order derivative calculation method for the temperature rise deviation change rate. The specific calculation steps are: first, perform numerical differentiation on the temperature rise deviation sequence data, calculate the difference between adjacent data points and divide by the time interval to obtain the first-order derivative sequence of the temperature rise deviation. Then perform numerical differentiation on the first-order derivative sequence again, calculate the difference between adjacent first-order derivative values ​​and divide by the time interval to obtain the second-order derivative sequence of the temperature rise deviation. The temperature rise curvature time series data refers to the data series of the second-order derivative of the temperature rise deviation changing with time, and the numerical unit is ℃ / s 2 , reflecting the acceleration characteristics of the temperature rise deviation change. The calculation process adopts the central difference format: the derivative value of the current point is equal to the difference between the values ​​of the previous and next two points divided by twice the time interval to ensure the accuracy of the calculation. The numerical calculation is processed in real time by the embedded computing unit, and the sliding window algorithm is used to reduce the amount of calculation. For example, within a 0.1 second time window, the temperature rise deviation changes from 0.2℃ to 0.6℃ and then to 0.3℃. The temperature rise curvature of this time period is negative, indicating that there is an inflection point change in the temperature rise deviation. The second-order difference curvature calculation can sensitively identify the local mutation and inflection point characteristics of the temperature rise deviation, and more accurately capture the shear heat layer disturbance mode caused by blade geometry anomalies than the first-order difference.

[0103] Speed-related threshold determination processing and coordinate mapping processing achieve accurate positioning of abnormal areas through adaptive threshold algorithms and geometric coordinate transformation. Speed-related threshold determination processing refers to an adaptive processing method that dynamically adjusts the temperature rise curvature determination threshold according to the real-time blade speed. The specific implementation process is as follows: First, a speed-threshold mapping relationship is established based on the blade speed time series data, and the threshold is set to ±0.5℃ / s at low speeds. 2 , the threshold is set to ±0.8℃ / s at medium speed 2 , the threshold is set to ±1.2℃ / s at high speed 2 The temperature rise curvature time series data is then compared with the dynamic threshold corresponding to the speed. When the absolute value of the temperature rise curvature exceeds the dynamic threshold, it is marked as an out-of-limit point. The temperature rise curvature out-of-limit area data refers to a data set consisting of all out-of-limit points, their timestamps, and spatial location information. Coordinate mapping processing utilizes a spatiotemporal coordinate transformation algorithm: the blade angular position is calculated based on the timestamp of the out-of-limit point and the blade speed. The axial position of the out-of-limit point in the flow channel is determined based on the axial coordinate data of the flow channel wall. The flow channel coordinates are then mapped to the blade surface coordinates using a geometric projection relationship. The axial coordinate data of the flow channel wall refers to the position coordinates of each sampling point along the axial direction of the flow channel, expressed in meters with respect to the flow channel inlet as the origin. The anomaly annular zone coordinate data contains three parameters: the chord percentage represents the relative position of the anomaly location in the blade chord direction, the radial percentage represents the relative position of the anomaly location in the blade radial direction, and the timestamp records the time when the anomaly occurred. For example, if an out-of-limit point occurs at a blade angle of 120 degrees and a flow channel axial position of 1.5 meters, the mapped anomaly annular zone coordinates have a chord percentage of 60% and a radial percentage of 75%. This speed-adaptive threshold design fully considers the changing pattern of the shear strength of the blade at different speeds, avoiding the problem of the fixed threshold being overly sensitive at low speeds and insufficiently sensitive at high speeds, ensuring the accuracy and consistency of anomaly detection.

[0104] In one embodiment of the present invention, the speed-related threshold determination processing is performed based on the temperature-rise curvature time series data and the blade speed time series data to obtain the temperature-rise curvature limit-exceeding region data; and the coordinate mapping processing is performed on the temperature-rise curvature limit-exceeding region data based on the axial coordinate data of the flow channel wall to obtain the abnormal annular zone coordinate data, including:

[0105] Performing time integration processing on the blade rotation speed time series data to obtain blade angular position time series data;

[0106] Performing speed-related threshold determination processing based on the temperature-rise curvature time series data and the blade angle position time series data to obtain temperature-rise curvature exceeding limit region data;

[0107] Performing polar coordinate mapping processing on the temperature rise curvature exceeding limit area data and the blade angular position time series data to obtain polar coordinate sequence data of the blade outer surface;

[0108] Circumferential stitching processing is performed on the polar coordinate sequence data of the blade outer surface according to the blade circumferential pitch data, and axial mapping processing is performed on the axial coordinate data of the flow channel wall to obtain abnormal annular zone coordinate data.

[0109] The following is a detailed description of the steps involved in the above embodiment:

[0110] Time integration converts speed time series data into angular position information using a numerical integration algorithm. Blade speed time series data contains the blade speed value at each moment, measured in revolutions per minute (RPM), with a sampling frequency of 2000 Hz. Time integration involves accumulating speed data over time to obtain the blade's cumulative rotation angle. The specific calculation process is as follows: First, the speed data is converted from RPM to radians per second (rad / s), using a conversion factor of 2π / 60. The converted angular velocity data is then numerically integrated using the trapezoidal integration method: the angular velocity values ​​at two adjacent moments are summed and divided by 2 to obtain the average angular velocity. This is then multiplied by the time interval of 0.0005 seconds to obtain the angle increment for that time period. The angle increments for all time periods are then accumulated to obtain the cumulative angle from the integration start time to the current moment. Blade angular position time series data refers to the cumulative angle of the blade relative to its initial position over time, measured in radians, with a periodic range of 0 to 2π. Double-precision floating-point arithmetic is used to ensure integration accuracy, with cumulative error kept within 0.01 radians. For example, when a blade rotates at a constant speed of 600 rpm, it rotates 2π radians per second, and the angular position data obtained by integration calculation shows a linear growth trend. This high-precision angular position calculation provides an accurate time-angle correspondence for subsequent spatial coordinate mapping, ensuring that anomaly detection results can accurately locate the specific angular position of the blade.

[0111] The speed-related threshold determination process uses a dynamic threshold algorithm and logical judgment to accurately identify temperature rise anomalies. This processing method determines the current speed state of the blade based on the blade angular position timing data and dynamically adjusts the determination threshold of the temperature rise curvature. The specific implementation process is: first calculate the time derivative of the blade angular position timing data to obtain the real-time angular velocity, and convert the angular velocity into the speed unit rpm. Then determine the corresponding threshold range based on the speed value: when the speed is 300-600 rpm, the threshold is set to ±0.4℃ / s 2 , when the speed is 600-900 rpm, the threshold is set to ±0.7℃ / s 2 , when the speed is 900-1200 rpm, the threshold is set to ±1.0℃ / s 2. Next, the temperature rise curvature time series data is compared with the dynamic threshold point by point: when the absolute value of the temperature rise curvature exceeds the threshold of the corresponding speed, the moment is marked as an over-limit point. The temperature rise curvature over-limit area data refers to a data set consisting of the timestamps, temperature rise curvature values, and corresponding angular positions of all over-limit points. The judgment algorithm is executed in real time using a programmable logic controller (PLC), and the response time is less than 1 millisecond. For example, at a blade angular position of π / 2 radians and a speed of 800 rpm, if the temperature rise curvature is -0.9°C / s 2 , exceeds the threshold corresponding to the speed ±0.7℃ / s 2 , the point is marked as an out-of-limit area. The speed segmentation threshold design fully considers the changing patterns of shear strength and thermal diffusion characteristics of the blade at different speeds. At low speeds, where shear is weak, a smaller threshold is set to increase sensitivity. At high speeds, where shear is stronger, a larger threshold is set to avoid misjudgment, ensuring the consistency and accuracy of anomaly detection across the entire speed range.

[0112] Polar coordinate mapping converts flow channel inspection data into blade surface coordinates using a coordinate transformation algorithm. Polar coordinate mapping involves converting rectangular coordinate data into a polar coordinate system with the blade center as the origin. The specific transformation process is as follows: First, the blade center axis is determined as the polar coordinate origin, and the blade angular position time series data is used directly as the polar coordinates. Then, based on the timestamp information in the temperature rise curvature limit violation area data, the corresponding angular position value in the blade angular position time series data is searched as the polar angle. Next, the polar diameter is determined based on the radial position of the violation point in the flow channel: the flow channel inner diameter is the blade root radius of 0.1 meters, and the flow channel outer diameter is the blade tip radius of 0.3 meters. The radial position of the violation point is determined by the radial distribution of sensors on the flow channel wall. The blade outer surface polar coordinate sequence data is the polar coordinate representation of all violation points on the blade surface, containing two parameters: polar angle and polar diameter. The polar angle is expressed in radians, ranging from 0 to 2π; the polar diameter is expressed in meters, ranging from 0.1 to 0.3 meters. Coordinate transformations are implemented using mathematical library functions to ensure accuracy and computational efficiency. For example, if an overrun point occurs at a blade angle of π radians and a radial position of 0.2 meters in the flow channel, the polar coordinates of that point are (π, 0.2). This polar coordinate representation naturally matches the rotational geometry of the blade, providing the most suitable coordinate system for subsequent spatial position processing.

[0113] Circumferential stitching and axial mapping processes achieve complete location of the anomaly ring through geometric stitching algorithms and coordinate transformation. Blade circumferential pitch data refers to the angular spacing between adjacent characteristic points along the circumferential direction of the blade. For multi-blade wind turbines, it represents the angular range occupied by a single blade, expressed in radians. Circumferential stitching involves normalizing the polar coordinate data to the angular range based on the blade's geometric characteristics. The specific implementation process is as follows: First, the angular range of a single blade is determined based on the blade circumferential pitch data. For example, the circumferential pitch of a five-blade wind turbine is 2π / 5 radians. A modulo operation is then performed on the polar coordinates in the polar coordinate sequence data on the blade's outer surface, mapping any angles exceeding the angular range back to zero within the pitch range. Next, axial mapping is performed: The axial position of the out-of-limit point in the flow channel is determined based on the axial coordinate data on the flow channel wall. These axial coordinates are then mapped to relative axial positions on the blade surface using geometric projection. Axial mapping utilizes a linear scaling transformation: the flow channel inlet corresponds to the leading edge of the blade, the flow channel outlet corresponds to the trailing edge, and intermediate positions are calculated using proportional interpolation. The coordinate data for the anomaly ring includes three parameters: the chord percentage represents the normalized position of the anomaly along the blade chord length, with a range of 0-100%; the radial percentage represents the normalized position of the anomaly along the blade radial direction, with a range of 0-100%; and the timestamp records the specific moment of anomaly detection. For example, if an out-of-limit point is located at the axial midpoint of the flow channel, the 75% radial position, and the blade rotation angle π / 5 radians, the coordinates of the anomaly ring with a chord percentage of 50% and a radial percentage of 75% are obtained after processing. This complete coordinate mapping system enables precise conversion from flow channel detection coordinates to blade surface coordinates, providing an accurate spatial positioning foundation for subsequent submillimeter wave precision projection and acoustic sampling.

[0114] Please continue reading Figure 1 , determining the oblique projection direction of the submillimeter wave according to the coordinate data of the abnormal ring zone, performing refraction phase analysis processing on the echo signal after the submillimeter wave passes through the abnormal ring zone, and obtaining a warping fingerprint parameter set;

[0115] In one embodiment of the present invention, the submillimeter wave oblique projection direction is determined according to the coordinate data of the abnormal annulus, and a refraction phase analysis process is performed on the echo signal after the submillimeter wave passes through the abnormal annulus to obtain a warping fingerprint parameter set, including:

[0116] Performing oblique projection angle band calculation processing based on the abnormal annular zone coordinate data and the blade radial dielectric constant distribution data to obtain submillimeter wave oblique projection direction data;

[0117] Performing oblique transmission control processing on the submillimeter wave beam according to the submillimeter wave oblique projection direction data to obtain submillimeter wave echo original time series data;

[0118] performing time normalization processing on the submillimeter wave echo original time series data according to the blade angle position time series data to obtain submillimeter wave echo normalized sequence data;

[0119] performing refraction phase unwrapping processing on the submillimeter wave echo normalized sequence data to obtain refraction phase sequence data;

[0120] Peak extraction, tail attenuation estimation, phase slope calculation and oscillation period determination processing are performed on the refraction phase sequence data to obtain a warping fingerprint parameter set.

[0121] The following is a detailed description of the steps involved in the above embodiment:

[0122] The oblique projection angle band calculation process uses electromagnetic wave propagation theory and optimization algorithms to determine the optimal submillimeter wave transmission parameters. The radial dielectric constant distribution data for blades refers to the distribution of dielectric constant values ​​at various radial locations from the root to the tip of the blade material. This data is obtained by measuring the dielectric properties of blade samples in the submillimeter wave band using a dielectric constant tester. For wind turbine blades made of aluminum alloy, the dielectric constant is approximately 8.5-9.2, while for plastic composite materials, it is approximately 3.2-4.1. The oblique projection angle band calculation process uses a numerical optimization method to calculate the optimal submillimeter wave projection angle and frequency range based on the spatial location of the anomaly band and the dielectric properties of the material. The specific calculation process is as follows: first, the radial location of the target area on the blade is determined based on the anomaly band coordinate data. The dielectric constant value at the corresponding location is extracted from the radial dielectric constant distribution data. The wavelength and penetration depth of the submillimeter wave at this dielectric constant are then calculated: the penetration depth is inversely proportional to the square root of the frequency and the square root of the dielectric constant. Next, angle optimization calculations are performed: taking the center of the anomaly annulus as the target point, the geometric angle from the submillimeter wave transmitter to the target point is calculated, taking into account the influence of the blade surface curvature on the beam propagation path, and determining the oblique projection angle with minimum reflection loss. The frequency band selection uses a penetration depth constraint: a frequency range with a penetration depth of 1.5-2 times the blade thickness is selected to ensure that the signal can completely penetrate the anomaly area. The submillimeter wave oblique projection direction data includes two parameters: the projection angle and the operating frequency. The projection angle is measured in degrees, with a range of 15-75 degrees; the operating frequency is measured in GHz, with a range of 100-300GHz. For example, for the anomaly area of ​​an aluminum alloy blade with a radial position of 70% and a dielectric constant of 8.8, the optimal projection angle is calculated to be 45 degrees and the operating frequency is 150GHz. This parameter optimization based on material properties ensures that the submillimeter wave can pass through the anomaly annulus under the best propagation conditions, achieving the highest signal-to-noise ratio and measurement accuracy.

[0123] The oblique transmission control process achieves precise submillimeter wave projection through controllable beamforming technology and precise mechanical positioning. The submillimeter wave transmitter uses a traveling wave tube oscillator to generate electromagnetic waves in the 100-300 GHz frequency band, with an output power of 10-50 mW. Oblique transmission control refers to a method for controlling the submillimeter wave transmitter's angle, frequency, power, and other parameters based on calculated projection parameters to achieve precise beam projection. The specific control process is as follows: First, based on the projection angle parameter in the submillimeter wave oblique projection direction data, a stepper motor drives the transmitter to the specified angle, with an angle control accuracy of ±0.1 degrees. The oscillator's output frequency is then adjusted based on the operating frequency parameter, achieving a frequency stability of ±0.01%. Next, beam transmission is initiated with a transmission time of 100 milliseconds, and the transmission power is automatically adjusted based on the target distance. Simultaneously, the receiver system is activated: the submillimeter wave receiver uses a Schottky diode detector with a response time of less than 1 microsecond and a reception bandwidth covering the transmission frequency range. Submillimeter wave echo raw time series data refers to the time-varying voltage sequence of the reflected signal collected by the receiver. The sampling frequency is 10 MHz, and the sampling duration is twice the transmission time, or 200 milliseconds. The data acquisition system uses a high-speed ADC with 16-bit resolution and 80 dB dynamic range. For example, when projecting a 150 GHz submillimeter wave at a 45-degree angle, the receiver begins to receive the reflected signal 15 microseconds after transmission. The signal strength shows a pattern of increasing and then decreasing over time. The completely dark environment eliminates interference from visible light on the electromagnetic wave receiver, allowing accurate detection of the weak reflected signal.

[0124] Time normalization eliminates the effects of blade rotation on signal timing through time base correction and interpolation algorithms. Time normalization converts time-varying signals acquired while the blade is rotating into signals processed using a time base relative to a fixed blade position. The specific process is as follows: First, the blade's angular position and angular velocity at the time of submillimeter wave transmission are calculated based on the blade's angular position time series data. Next, the submillimeter wave propagation time is calculated: based on the distance from the transmitter to the anomaly ring and the electromagnetic wave propagation velocity, the one-way propagation time is estimated to be approximately 0.5-2 microseconds. Next, a time base conversion is performed: the time axis of the raw submillimeter wave echo time series data is converted from the actual acquisition time to relative time relative to a fixed point on the blade surface. This time conversion utilizes a linear interpolation algorithm: a time offset is calculated based on the blade's angular velocity, and the raw data is resampled and interpolated. The normalized submillimeter wave echo sequence data is the reflection signal data sequence that has undergone time base correction, eliminating the Doppler effect and time offset caused by blade rotation. Data processing utilizes a digital signal processor (DSP) for real-time computation, with processing latency less than 10 milliseconds. For example, when a blade rotates at 1000 rpm, it rotates approximately 1.67 degrees within a 100-millisecond acquisition period. Time normalization corrects this rotation-induced time variation to zero, making signal feature extraction unaffected by rotation speed. This unified time base ensures comparability of reflection signals acquired at different rotation speeds, providing a stable data foundation for subsequent phase analysis.

[0125] Refraction phase unwrapping uses a phase unwrapping algorithm to extract submillimeter wave phase variation information. This process involves unwrapping a periodic phase signal into a continuously varying phase function. When submillimeter waves pass through the shear heat layer on the blade surface and the air interface, differences in dielectric constants cause phase delays. This delay information is incorporated into the phase of the reflected signal. The specific processing steps are as follows: First, a Hilbert transform is performed on the normalized submillimeter wave echo sequence data to extract the instantaneous phase of the signal. Then, a phase unwrapping algorithm is applied to detect 2π jumps in the phase data and adjust phase differences exceeding π to corresponding continuous phase values. Phase unwrapping utilizes a quality-guided approach, prioritizing processing in areas of high signal quality and gradually expanding to areas of lower quality. The refraction phase sequence data, expressed in radians, represents the continuous phase variation data after unwrapping. These values ​​reflect the cumulative phase variation of the submillimeter wave as it passes through the anomaly ring. Phase data accuracy reaches 0.01 radians, corresponding to a ranging accuracy of approximately tens of microns. For example, when a submillimeter wave passes through a 1 mm thick shear-heat layer, the refractive index change caused by the temperature difference produces a phase delay of approximately 0.5 radians. This high-precision phase information directly reflects the geometric deformation and density distribution anomalies in the anomalous ring, providing sensitive physical parameters for quantitative analysis of warping characteristics.

[0126] The characteristic parameter extraction process uses multiple signal analysis algorithms to extract quantitative characteristics of blade warpage from phase data. Peak extraction uses a local extremum search algorithm to locate the maximum phase variation within the refraction phase sequence data. This peak reflects the maximum phase delay within the anomaly. Tail attenuation estimation uses an exponential fitting method to analyze the decay process after the phase peak: the data segment after the peak is selected and fitted with an exponential decay function using the least squares method to extract the decay time constant. Phase slope calculation uses numerical differentiation: the first-order derivative of the phase data is calculated, and the maximum phase change rate is extracted as the slope feature. Oscillation period determination is achieved through autocorrelation analysis: the autocorrelation function of the phase data is calculated, and the time delay corresponding to the first positive peak is found as the oscillation period. The warpage fingerprint parameter set consists of four dimensionless parameters: the refraction peak represents the phase delay intensity of the anomaly, the echo tail represents the phase attenuation characteristics, the slope change rate represents the phase gradient characteristics, and the phase oscillation period represents the periodicity of the phase fluctuations. The values ​​of these four parameters are normalized to a dimensionless range between 0 and 1. For example, the warping fingerprint parameters for a particular anomaly ring are [0.65, 0.42, 0.78, 0.33], characterizing the specific warping characteristics at that location. This multi-parameter feature extraction method compresses complex phase information into four key indicators, preserving the key information of the anomaly while facilitating subsequent numerical comparisons and cross-validation, achieving precise quantitative characterization of blade micro-warping.

[0127] In one embodiment of the present invention, performing oblique projection angle band calculation processing based on the abnormal annular zone coordinate data and the blade radial dielectric constant distribution data to obtain submillimeter wave oblique projection direction data includes:

[0128] Performing coordinate matching processing on blade radial thickness distribution data according to the abnormal annular zone coordinate data to obtain annular zone thickness sequence data;

[0129] Performing attenuation compensation calculation processing based on the annular belt thickness sequence data and the blade radial dielectric constant distribution data to obtain annular belt effective penetration depth data;

[0130] Performing bandwidth screening processing on the available submillimeter wave spectrum interval according to the annular band effective penetration depth data to obtain annular band candidate frequency band data;

[0131] Angle and frequency band joint optimization processing is performed according to the candidate frequency band data of the ring band and the abnormal coordinate data of the abnormal ring band to obtain submillimeter wave oblique projection direction data.

[0132] The following is a detailed description of the steps involved in the above embodiment:

[0133] Coordinate matching uses a spatial coordinate association algorithm to precisely match the location of the abnormal annulus with blade geometric parameters. Blade radial thickness distribution data refers to the distribution of blade thickness values ​​at various radial locations from root to tip. This data is obtained by scanning the blade surface with a three-dimensional coordinate measuring machine (CMM). The data format is an array of radial position percentages and corresponding thickness values. Coordinate matching involves extracting the corresponding geometric parameters from a blade geometry database based on the spatial location of the abnormal annulus. The specific implementation process involves first extracting a radial percentage parameter from the abnormal annulus coordinate data. This parameter represents the relative position of the abnormal location in the blade radial direction, with a value range of 0-100%. Then, the radial percentage is used to index the position in the blade radial thickness distribution data, and a linear interpolation algorithm is used to calculate the thickness value at the precise location. Next, the chord length percentage parameter is used to determine the chord-wise location of the abnormal annulus. The blade thickness at that chord-wise location is calculated using the blade's three-dimensional geometric model. The annular thickness series data refers to the series of blade thickness values ​​at each sampling point within the area covered by the abnormal annulus, with a sampling interval of 1 mm and data units in millimeters. Data processing is performed using a CAD geometry calculation engine, achieving a precision of 0.01 mm. For example, if an abnormal annulus is located at 75% of the radial direction and 60% of the chord length, coordinate matching reveals that the blade thickness in this area is 3.2 mm and the annulus length is 8 mm, resulting in a sequence of eight thickness values: [3.2, 3.1, 3.3, 3.2, 3.0, 3.1, 3.2, 3.1]. This precise geometric matching ensures the accuracy of the geometric parameters used in subsequent electromagnetic wave propagation calculations and avoids the calculation errors that may be caused by using average thickness.

[0134] The attenuation compensation calculation process uses electromagnetic wave propagation theory to calculate the effective penetration of submillimeter waves in blade materials. This process is an electromagnetic calculation method that calculates electromagnetic wave penetration loss and determines the effective penetration depth based on the dielectric properties and thickness distribution of the blade material. The specific calculation process is as follows: First, the material thickness required for submillimeter waves to penetrate is determined based on the annular band thickness series data. The dielectric constant and loss tangent values ​​at the corresponding locations are then extracted from the blade radial dielectric constant distribution data. Next, the attenuation constant is calculated: the attenuation constant is proportional to the square root of the frequency, the square root of the dielectric constant, and the loss tangent. The penetration depth is then calculated: the penetration depth is defined as the propagation distance at which the electromagnetic wave power decays to 1 / e of its initial value, and is equal to the inverse of the attenuation constant. The annular band effective penetration depth data is a series of values ​​indicating the maximum penetration depth of submillimeter waves at each location in the anomalous annular band, expressed in millimeters. The calculation takes into account the frequency dispersion characteristics of the blade material: the dielectric constant variation rate of aluminum alloys in the 100-300 GHz frequency band is approximately 5%, while that of plastic composites is approximately 15%. For example, for an aluminum alloy blade with a thickness of 3.2 mm and a dielectric constant of 8.8, the effective penetration depth at 150 GHz is 4.8 mm, which is greater than the blade thickness, indicating that this frequency can fully penetrate. However, at 250 GHz, the effective penetration depth is only 2.1 mm, less than the blade thickness, resulting in severe signal attenuation. This material-property-based penetration depth calculation avoids the problem of excessive signal attenuation caused by blindly selecting an excessively high frequency, ensuring that submillimeter waves can achieve sufficient signal strength for accurate measurements.

[0135] Bandwidth screening uses frequency domain response analysis to determine the frequency range that meets penetration requirements. The available submillimeter wave spectrum range refers to the submillimeter wave frequency range supported by the system hardware. In this solution, it is 100-300 GHz, determined by the operating bandwidth of the traveling wave tube oscillator. Bandwidth screening involves selecting suitable operating frequencies from the available spectrum based on penetration depth requirements. The specific screening process involves first setting a penetration depth criterion, requiring the effective penetration depth to be greater than 1.5 times the maximum thickness value in the annular band thickness sequence data to ensure that the signal can penetrate the entire anomaly area. Then, point-by-point calculations are performed within the 100-300 GHz spectrum range in 10 GHz steps to assess whether the penetration depth meets the requirements at each frequency. Next, signal-to-noise ratio optimization is performed: among the frequencies that meet the penetration requirements, the frequency range with the highest reflection coefficient is selected. The reflection coefficient is proportional to the square of the dielectric constant difference. The candidate annular band data refers to the frequency range data that has been screened for both penetration depth and signal-to-noise ratio. It includes three parameters: start frequency, end frequency, and center frequency, and the data is expressed in GHz. The screening algorithm utilizes a multi-objective optimization approach: penetration depth is weighted 60%, and signal-to-noise ratio is weighted 40%. The frequency band with the highest overall score is selected as a candidate. For example, for an anomaly with a maximum thickness of 3.3 mm, the required penetration depth is greater than 4.95 mm. After calculation, the 130-180 GHz frequency band, with a center frequency of 155 GHz and a bandwidth of 50 GHz, meets this requirement. This multi-criteria frequency band screening ensures signal penetration while optimizing the measurement signal-to-noise ratio, avoiding the performance bias that can arise from a single criterion.

[0136] The joint optimization of angle and frequency bands determines the optimal submillimeter wave operating parameters through multi-parameter collaborative optimization. Joint optimization refers to a multi-objective optimization method that simultaneously optimizes both the projection angle and the operating frequency to achieve optimal performance for the submillimeter wave detection system. The specific optimization process involves constructing an optimization objective function, which includes three evaluation metrics: signal strength, geometric coverage, and measurement accuracy. Signal strength is proportional to the cosine of the projection angle and inversely proportional to the square root of the frequency; geometric coverage exhibits a nonlinear relationship with the projection angle, reaching its optimum around 45 degrees; and measurement accuracy is proportional to the frequency and inversely proportional to the projection path length. Geometric constraints are then determined based on the coordinate data of the anomaly ring: the projection angle is constrained by the anomaly ring location and the transmitter installation position, with an angle range of 15-75 degrees; and the operating frequency is constrained by the candidate frequency band data for the ring. A genetic algorithm is then used to perform multi-objective optimization: the population size is set to 50, the number of evolution generations is set to 100, the crossover probability is set to 0.8, and the mutation probability is set to 0.1. The optimization algorithm searches for the optimal solution within the feasible region, with the convergence criterion being that the optimal solution remains unchanged by less than 1% for 10 consecutive generations. Submillimeter wave oblique projection data includes two parameters: the optimal projection angle and the optimal operating frequency. The projection angle accuracy is ±0.5 degrees, and the operating frequency accuracy is ±1 GHz. For example, for an anomaly located at 75% of the radial direction and 60% of the chord length, the candidate frequency band is 130-180 GHz. After joint optimization, the optimal projection angle is 42 degrees and the optimal operating frequency is 158 GHz. The overall performance score of this parameter combination is approximately 25% higher than that of optimizing any parameter alone. This multi-parameter joint optimization avoids the local optimality that may result from optimizing either the projection angle or the operating frequency alone, ensuring the overall performance optimization of the submillimeter wave detection system under complex geometric conditions.

[0137] Please continue reading Figure 1 According to the abnormal annular coordinate data, acoustic sampling of the corresponding area is activated to obtain the air-acoustic resonance standing wave data generated by the blade rotation. The air-acoustic resonance standing wave data is cross-validated with the warping fingerprint parameter set to obtain the quantitative detection and judgment results of the blade geometric defects.

[0138] In one embodiment of the present invention, activating acoustic sampling of the corresponding area according to the abnormal annular zone coordinate data to obtain air-acoustic resonance standing wave data generated by blade rotation, cross-validating the air-acoustic resonance standing wave data with the warping fingerprint parameter set to obtain a quantitative detection and determination result of blade geometric defects includes:

[0139] Determining the acoustic sampling window coordinate data according to the abnormal ring coordinate data;

[0140] Acquiring air-acoustic resonance standing wave time series data according to the acoustic sampling window coordinate data to obtain air-acoustic resonance standing wave data;

[0141] performing rotation order demodulation processing on the air-acoustic resonance standing wave data to obtain order energy spectrum data;

[0142] Performing amplitude-phase mapping processing on the order energy spectrum data according to the blade pitch angle distribution data to obtain pitch angle mapping spectrum data;

[0143] A feature coupling judgment process is performed based on the pitch angle mapping spectrum data and the warping fingerprint parameter set to obtain a quantitative detection and judgment result of the blade geometric defect.

[0144] The following is a detailed description of the steps involved in the above embodiment:

[0145] The acoustic sampling window coordinate data determines the location of the anomaly annulus and converts it into the activation area of ​​the acoustic sensor array through a spatial mapping algorithm. The acoustic sampling window coordinate data refers to the coordinate information of the sensor locations in the acoustic sensor array on the channel wall that need to be activated. This information is used to accurately acquire acoustic signals from the area corresponding to the anomaly annulus. The specific implementation process is as follows: First, based on the chord length percentage and radial percentage parameters in the anomaly annulus coordinate data, a geometric projection algorithm is used to calculate the projection position of the anomaly annulus on the channel wall. The channel wall is arranged with a 10×8 matrix of MEMS acoustic sensors, with a 5 cm spacing between sensors, covering the entire channel wall area. The corresponding sensor number is then determined based on the projection position. The projection area of ​​the anomaly annulus is expanded to a square sampling window with a side length of 10 cm to ensure complete coverage of the spatial range of the anomaly signal. Next, the number and specific locations of sensors within the sampling window are calculated. The sampling window contains four sensors in a 2×2 subarray, and the row and column coordinates of each sensor are recorded. The acoustic sampling window coordinate data includes four parameters: the number of activated sensors, the row coordinates, the column coordinates, and the sampling window center coordinates. For example, if an abnormal annulus is projected onto the channel wall at row 5, column 6, the sampling window is determined to cover four sensors, rows 5-6 and columns 6-7, with the sampling window's center coordinates at (5.5, 6.5). This precise spatial positioning avoids the potential background noise interference of activating the entire sensor array, restricting the acoustic acquisition range to the vicinity of the abnormal area, significantly improving the signal-to-noise ratio and detection accuracy.

[0146] Acquisition of standing wave data for aeroacoustic resonance (AAR) accurately captures acoustic signals in abnormal areas through selective sensor activation and high-frequency data acquisition. AAR refers to the physical phenomenon of periodic airflow disturbances generated by blade rotation resonating with the acoustic cavity of the flow channel. This geometric disturbance generates a characteristic standing wave pattern in the annular zone. The acquisition process is as follows: First, the MEMS acoustic sensor at the corresponding location is activated based on the acoustic sampling window coordinate data, while other sensors outside the sampling window are disabled to reduce background noise. Data acquisition parameters are then set: a sampling frequency of 20 kHz, covering the AAR frequency range of 2.5-3.8 kHz, an acquisition time of 2 seconds, and 40,000 data points. Next, data acquisition is initiated synchronously: the acquisition system is synchronized with the blade rotary encoder signal to ensure accurate temporal correspondence between the acoustic data and the blade angular position. AAR standing wave time series data refers to the time-varying numerical sequence of the sound pressure signal acquired by the activated sensor, measured in Pascals, reflecting the acoustic response characteristics of the annular zone. Data preprocessing includes DC component removal and anti-aliasing filtering: a high-pass filter is used to remove low-frequency noise below 100Hz, and a low-pass filter is used to remove high-frequency interference above 10kHz. Air-acoustic resonance standing wave data is a characteristic parameter obtained through statistical analysis of time series data. It includes three key indicators: frequency domain energy distribution, coherence function, and standing wave node location. For example, at a blade speed of 1000 rpm, the main frequency of air-acoustic resonance generated by the abnormal annulus is 3.2kHz, the standing wave node spacing is 5.3cm, and the sound pressure amplitude is 15dB higher than that in the normal area. The completely dark environment eliminates the optical noise interference of the acoustic sensor, allowing the weak air-acoustic resonance signal to be accurately detected.

[0147] Rotational order demodulation extracts acoustic characteristics related to blade rotation through angle-domain resampling and order analysis. Rotational order demodulation is a signal processing method that converts time-domain acoustic signals into angle-domain signals and then performs spectral decomposition based on blade rotation orders. Rotational order refers to the relationship between the acoustic signal frequency and the blade rotation frequency as integer multiples, reflecting the degree of correlation between acoustic phenomena and blade geometric characteristics. The specific processing steps are as follows: First, the cumulative rotation angle is calculated based on the time-series blade speed data. The aeroacoustic resonance standing wave data is resampled from the time domain to the angle domain to eliminate the influence of speed fluctuations on the spectral analysis. Next, order tracking is performed: the angle-domain signal is resampled at equal angles based on the blade rotation angle, with a sampling resolution of 0.1 degrees, obtaining 3600 data points per rotation. Next, order spectrum analysis is performed: a fast Fourier transform is performed on the angle-domain signal to calculate the energy distribution of each order. The order energy spectrum data represents the distribution of acoustic energy corresponding to each rotation order, with the order number on the horizontal axis and the energy density on the vertical axis, with data units expressed in decibels. Order analysis covers the 0.5-20 order range with a resolution of 0.1 order. For example, a five-blade fan exhibits an energy peak at fundamental frequency order 5, while an abnormal blade exhibits an additional energy peak at order 5.2. This peak is 8dB lower than the normal peak but significantly higher than the background noise. This order-domain analysis effectively separates the acoustic components related to blade rotation from random noise, providing a high-purity rotation-related signal for subsequent feature coupling analysis.

[0148] Amplitude-phase mapping achieves a geometric interpretation of acoustic characteristics through correlation analysis between blade geometric parameters and order energy data. Blade pitch angle distribution data refers to the distribution of spiral angles at various radial locations on the blade, characterizing the blade's torsional geometric characteristics. This data is obtained by scanning with a three-dimensional coordinate measuring machine (CMM) and is measured in degrees. Amplitude-phase mapping is a data association method that maps order energy spectrum data to the acoustic response distribution at corresponding geometric locations based on the blade pitch angle geometry. The specific mapping process is as follows: First, a correspondence between pitch angle and order frequency is established. Positions with larger pitch angles produce higher-order acoustic responses. The mapping coefficient is the pitch angle divided by 360 degrees and multiplied by the number of blades. Then, based on the energy values ​​of each order in the order energy spectrum data, an inverse mapping algorithm is used to calculate the acoustic response intensity at the corresponding pitch angle position. Next, amplitude and phase separation is performed: amplitude and phase information are extracted from the complex order data, respectively. The amplitude reflects the acoustic energy intensity, while the phase reflects the time delay characteristics of the acoustic response. The pitch angle mapping spectrum data refers to the acoustic response amplitude and phase distribution corresponding to each pitch angle position. The data format is a three-tuple array of pitch angle, response amplitude, and response phase. The mapping resolution is 1 degree pitch angle, covering the pitch angle variation range of 0-45 degrees of the blade. For example, the mapping spectrum at the pitch angle of 25 degrees shows a response amplitude of -12dB and a response phase of 135 degrees, indicating that there is an acoustic anomaly at this position relative to the normal geometry. This geometric-acoustic mapping establishes a direct correlation between the blade's geometric characteristics and the acoustic response, allowing the acoustic detection results to accurately correspond to the specific geometric position of the blade.

[0149] Feature-coupling judgment processing uses a multi-physics cross-validation algorithm to achieve comprehensive blade defect determination. This method involves performing a multi-dimensional comparative analysis of the pitch angle mapping spectrum data and the warping fingerprint parameter set, confirming the presence of defects through consistency verification. The specific judgment process begins by establishing feature correspondences: the response amplitude in the pitch angle mapping spectrum data corresponds to the refraction peak in the warping fingerprint parameter set, and the response phase corresponds to the phase oscillation period, establishing a one-to-one correspondence between the four pairs of feature parameters. Numerical normalization is then performed: both data sets are normalized to the range of 0-1 to eliminate the influence of dimensional differences. Feature correlation is then calculated: the Pearson correlation coefficient is used to analyze the linear correlation between the two sets of feature parameters. A correlation coefficient greater than 0.7 indicates strong correlation, 0.4-0.7 indicates moderate correlation, and less than 0.4 indicates weak correlation or no correlation. A threshold judgment is then performed: a defect is determined when the correlation coefficients of all four pairs of feature parameters are greater than 0.6, and at least two pairs are greater than 0.8. The quantitative detection and judgment results of blade geometric defects include three output parameters: defect determination conclusion, confidence score, and defect type identification. The defect determination conclusion is a binary output: 1 indicates the presence of a defect, 0 indicates the absence of a defect; the confidence score is a continuous value ranging from 0-100%, reflecting the reliability of the determination result; defect type identification distinguishes between three defect modes: warping, mass shift, and composite. For example, the characteristic coupling analysis of a blade showed that the correlation coefficients of four pairs of parameters were 0.85, 0.78, 0.92, and 0.71, respectively, all exceeding the threshold requirements. The determination result was that a defect existed, with a confidence level of 89%, and the defect type was warping. This multi-physical quantity cross-validation mechanism effectively avoids the misjudgment that may occur with a single detection method. By confirming the consistency of two different physical principles, acoustics and electromagnetics, it significantly improves the accuracy and reliability of defect detection.

[0150] The present invention also provides an unmanned intelligent production line for automobile fans. Since the unmanned intelligent production line for automobile fans adopts the unmanned intelligent production method for automobile fans of any of the above-mentioned embodiments, it has the advantages of any of the above-mentioned embodiments, which will not be repeated here.

[0151] The above description is only a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural transformation made by using the contents of the present description and drawings under the inventive concept of the present invention, or directly / indirectly applied in other related technical fields, is included in the patent protection scope of the present invention.

Claims

1. An unmanned intelligent production method for automobile fans, characterized in that: include: Perform low-speed rotation on the rotating blades entering the constant-pressure flow channel, continuously obtain the time series data of the shear heat layer temperature rise on the flow channel wall, and obtain the benchmark temperature rise curve; According to the reference temperature rise curve, differential processing is performed on the wall shear heat layer temperature rise time series data acquired in real time during the blade speed increase process, and the temperature rise curvature exceeding limit area is extracted and the abnormal annular band coordinate data is generated; specifically, the method includes: performing time base synchronization processing on the wall shear heat layer temperature rise time series data acquired in real time and the blade speed time series data to obtain temperature rise and speed synchronization sequence data; performing first-order differential processing on the temperature rise and speed synchronization sequence data according to the reference temperature rise curve to obtain temperature rise deviation sequence data; performing second-order differential curvature calculation processing on the temperature rise deviation sequence data to obtain temperature rise curvature time series data; performing speed-related threshold judgment processing on the temperature rise curvature time series data and the blade speed time series data to obtain temperature rise curvature exceeding limit area data, and performing coordinate mapping processing on the temperature rise curvature exceeding limit area data according to the axial coordinate data of the flow channel wall to obtain abnormal annular band coordinate data; Determine the oblique projection direction of the submillimeter wave according to the coordinate data of the abnormal ring zone, perform refraction phase analysis on the echo signal after the submillimeter wave passes through the abnormal ring zone, and obtain a warping fingerprint parameter set; According to the coordinate data of the abnormal annulus, the coordinate data of the acoustic sampling window of the corresponding area is activated to obtain the air-acoustic resonance standing wave data generated by the blade rotation, and the air-acoustic resonance standing wave data is cross-validated with the warping fingerprint parameter set to obtain the quantitative detection and judgment results of the blade geometric defects; wherein, the warping fingerprint parameter set includes the refraction peak, the echo tail, the slope change rate and the phase oscillation period, the refraction peak represents the phase delay intensity of the abnormal annulus, the echo tail represents the phase attenuation characteristics, the slope change rate represents the phase gradient characteristics, and the phase oscillation period represents the periodic characteristics of the phase fluctuation.

2. The unmanned intelligent production method for automobile fans according to claim 1, characterized in that: The low-speed rotation start process is performed on the rotating blades entering the constant-pressure flow channel, and the time series data of the shear heat layer temperature rise on the flow channel wall is continuously obtained to obtain a reference temperature rise curve, including: Performing segmented pressurization on the gas in the inlet, middle, and outlet sections of the flow channel to obtain axial pressure gradient distribution data; According to the axial pressure gradient distribution data, performing three-stage rotation speed increasing processing on the rotating blades respectively to obtain inlet section shear response data, middle section shear response data and outlet section shear response data; Performing real-time temperature rise sampling processing on the inlet section wall surface, the middle section wall surface and the outlet section wall surface to obtain the inlet section temperature rise time series data, the middle section temperature rise time series data and the outlet section temperature rise time series data; performing weighted fusion processing on the inlet section temperature rise time series data, the middle section temperature rise time series data, and the outlet section temperature rise time series data, respectively, based on the axial pressure gradient distribution data, the inlet section shear response data, the middle section shear response data, and the outlet section shear response data, to obtain axial comprehensive temperature rise time series data; An attenuation tail feature extraction process is performed on the axial comprehensive temperature rise time series data to obtain a reference temperature rise curve including a temperature rise peak value, an attenuation constant and a tail period.

3. The unmanned intelligent production method for automobile fans according to claim 2, characterized in that: The step of performing three-stage rotation speed increasing processing on the rotating blades according to the axial pressure gradient distribution data to obtain inlet section shear response data, middle section shear response data, and outlet section shear response data includes: Perform acceleration curve planning processing based on the blade radial moment of inertia distribution data and the blade radial natural vibration frequency data to obtain three-segment angular acceleration sequence data; Performing a first-stage speed increasing process according to the three-segment angular acceleration sequence data, acquiring inlet section shear response time series data when the blade speed reaches the inlet section target speed, and obtaining inlet section shear response data; Performing a second stage of speed increasing processing according to the three-segment angular acceleration sequence data, acquiring intermediate segment shear response time series data when the blade speed reaches the intermediate segment target speed, and obtaining intermediate segment shear response data; The third stage speed increasing processing is performed according to the three sections of angular acceleration sequence data, and when the blade speed reaches the outlet section target speed, the outlet section shear response time series data is acquired to obtain the outlet section shear response data.

4. The unmanned intelligent production method for automobile fans according to claim 2, characterized in that: The step of performing weighted fusion processing on the inlet section temperature rise time series data, the middle section temperature rise time series data, and the outlet section temperature rise time series data based on the axial pressure gradient distribution data, the inlet section shear response data, the middle section shear response data, and the outlet section shear response data, respectively, to obtain axial comprehensive temperature rise time series data, including: Normalizing the blade radial mass distribution data and the blade moment of inertia parameters, and combining them with the axial pressure gradient distribution data to obtain a segment weight basis matrix; performing response intensity weighting processing on the inlet segment shear response data, the middle segment shear response data, and the outlet segment shear response data according to the segment weight basic matrix to obtain a segment dynamic weight coefficient vector; According to the segment dynamic weight coefficient vector, multi-scale wavelet weighted fusion processing is performed on the inlet segment temperature rise time series data, the middle segment temperature rise time series data and the outlet segment temperature rise time series data to obtain axial comprehensive temperature rise time series data.

5. The unmanned intelligent production method for automobile fans according to claim 1, characterized in that: performing speed-related threshold determination processing based on the temperature-rise curvature time series data and the blade speed time series data to obtain temperature-rise curvature exceeding limit region data; The coordinate mapping process of the temperature rise curvature exceeding limit region data is performed according to the axial coordinate data of the flow channel wall to obtain the abnormal annular zone coordinate data, including: Performing time integration processing on the blade rotation speed time series data to obtain blade angular position time series data; performing speed-related threshold determination processing on the temperature-rise curvature time series data and the blade angle position time series data to obtain temperature-rise curvature exceeding limit region data; Performing polar coordinate mapping processing on the temperature rise curvature exceeding limit region data and the blade angular position time series data to obtain polar coordinate sequence data of the blade outer surface; Circumferential stitching processing is performed on the blade circumferential pitch data and the polar coordinate sequence data of the blade outer surface, and axial mapping processing is performed according to the axial coordinate data of the flow channel wall to obtain abnormal annular zone coordinate data.

6. The unmanned intelligent production method for automobile fans according to claim 1, characterized in that: The submillimeter wave oblique projection direction is determined according to the abnormal annular zone coordinate data, and a refraction phase analysis process is performed on the echo signal after the submillimeter wave passes through the abnormal annular zone to obtain a warping fingerprint parameter set, including: Performing oblique projection angle band calculation processing based on the abnormal annular zone coordinate data and the blade radial dielectric constant distribution data to obtain submillimeter wave oblique projection direction data; Performing oblique transmission control processing on the submillimeter wave beam according to the submillimeter wave oblique projection direction data to obtain submillimeter wave echo original time series data; performing time normalization processing on the submillimeter wave echo original time series data according to the blade angle position time series data to obtain submillimeter wave echo normalized sequence data; performing refraction phase unwrapping processing on the submillimeter wave echo normalized sequence data to obtain refraction phase sequence data; Peak extraction, tail attenuation estimation, phase slope calculation and oscillation period determination processing are performed on the refraction phase sequence data to obtain a warping fingerprint parameter set.

7. The unmanned intelligent production method for automobile fans according to claim 6, characterized in that: The performing of oblique projection angle band calculation processing based on the abnormal annular zone coordinate data and the blade radial dielectric constant distribution data to obtain submillimeter wave oblique projection direction data includes: Performing coordinate matching processing on blade radial thickness distribution data according to the abnormal annular zone coordinate data to obtain annular zone thickness sequence data; Performing attenuation compensation calculation processing based on the annular belt thickness sequence data and the blade radial dielectric constant distribution data to obtain annular belt effective penetration depth data; Performing bandwidth screening processing on the available submillimeter wave spectrum interval according to the annular band effective penetration depth data to obtain annular band candidate frequency band data; Angle and frequency band joint optimization processing is performed according to the candidate frequency band data of the ring band and the abnormal coordinate data of the abnormal ring band to obtain submillimeter wave oblique projection direction data.

8. The unmanned intelligent production method for automobile fans according to claim 1, characterized in that: The method activates the acoustic sampling window coordinate data of the corresponding area according to the abnormal annular zone coordinate data, obtains the air acoustic resonance standing wave data generated by the blade rotation, and cross-validates the air acoustic resonance standing wave data with the warping fingerprint parameter set to obtain a quantitative detection and judgment result of the blade geometric defect, including: Determining the acoustic sampling window coordinate data according to the abnormal ring coordinate data; Acquiring air-acoustic resonance standing wave time series data according to the acoustic sampling window coordinate data to obtain air-acoustic resonance standing wave data; performing rotation order demodulation processing on the air-acoustic resonance standing wave data to obtain order energy spectrum data; Performing amplitude-phase mapping processing on the order energy spectrum data according to the blade pitch angle distribution data to obtain pitch angle mapping spectrum data; A feature coupling judgment process is performed based on the pitch angle mapping spectrum data and the warping fingerprint parameter set to obtain a quantitative detection and judgment result of the blade geometric defect.

Images