A rapid detection device and method for evaluating aluminum alloy weld fatigue life

Abstract

The application provides a rapid detection device and method for evaluating aluminum alloy weld fatigue life, and relates to the technical field of fatigue detection. The device comprises: an ultrasonic excitation and collection module, which comprises a broadband power amplifier and a high-frequency piezoelectric probe, is used for contacting an aluminum alloy weld to be detected, and is configured to inject a group of high-energy and continuously swept-frequency ultrasonic excitation signals into the weld; a receiver, which is integrally integrated with the high-frequency piezoelectric probe, is used for synchronously collecting a first ultrasonic response signal modulated by internal micro-damage of the weld; a feature extraction module is configured to demodulate a high-order harmonic component caused by material fatigue damage from the first ultrasonic response signal; the relative energy ratio of the high-order harmonic component and a fundamental wave component is calculated, a nonlinear resonance spectrum distortion factor is generated; and a life evaluation module, which is embedded with a first fatigue damage physical model, is configured to take the nonlinear resonance spectrum distortion factor as an input parameter, calculate and output a first residual fatigue life of the aluminum alloy weld to be detected.

Description

Technical Field

[0001] This application relates to the field of fatigue testing technology, specifically to a rapid testing device and method for evaluating the fatigue life of aluminum alloy welds. Background Technology

[0002] In aerospace, rail transportation, and other fields, fatigue life assessment of aluminum alloy welded structures is a crucial step in ensuring the safe operation of equipment. Nonlinear ultrasonic testing technology, due to its high sensitivity to microscopic damage in materials, has become an effective means of early diagnosis of fatigue damage.

[0003] In the prior art, such as the invention patent with publication number CN104359977A: Characterization method of high-order nonlinear parameters of surface acoustic wave in bending fatigue state of metal sheet, a method is proposed to evaluate the bending fatigue state of metal sheet by detecting high-order harmonics of ultrasound.

[0004] However, this method employs a separate transmitter and receiver transducer design. Specifically, it requires a transducer with a center frequency of f to transmit surface acoustic waves, while another transducer with a center frequency of 2f is used specifically to receive the second harmonic signal. This discrete probe arrangement leads to difficulties in probe positioning, inconsistent sound wave propagation paths, and unstable coupling states during actual testing (especially in aluminum alloy weld areas with irregular surfaces). This introduces significant measurement errors, resulting in poor repeatability and low reliability of the test results, making it impossible to achieve rapid, stable, and repeatable on-site testing. Summary of the Invention

[0005] In view of the above-mentioned defects or deficiencies in the prior art, this application aims to provide a rapid testing device and method for evaluating the fatigue life of aluminum alloy welds, so as to achieve on-site testing with strong stability and high reliability.

[0006] In a first aspect, this application proposes a rapid detection device for evaluating the fatigue life of aluminum alloy welds, comprising an ultrasonic excitation and acquisition module, including:

[0007] Broadband power amplifier;

[0008] A high-frequency piezoelectric probe, connected to the broadband power amplifier, is used to contact the aluminum alloy weld to be tested and is configured to inject a set of high-energy, continuously sweeping ultrasonic excitation signals into the weld.

[0009] The receiver is integrated with the high-frequency piezoelectric probe and is used to synchronously acquire the first ultrasonic response signal modulated by the micro-damage inside the weld.

[0010] The feature extraction module, electrically connected to the ultrasonic excitation and acquisition module, is configured as follows:

[0011] Demodulate the higher-order harmonic components caused by material fatigue damage from the first ultrasonic response signal;

[0012] Calculate the relative energy ratio between the higher-order harmonic components and the fundamental component to generate the nonlinear resonance spectrum distortion factor;

[0013] The life assessment module has a first fatigue damage physical model embedded in it. The first fatigue damage physical model is configured to calculate and output the first remaining fatigue life of the aluminum alloy weld to be tested, using the nonlinear resonance spectrum distortion factor as the input parameter.

[0014] According to the technical solution provided in this application, the device also includes a calibration test block, the material and heat treatment state of which are the same as the initial non-service state of the aluminum alloy weld to be tested;

[0015] The feature extraction module is configured as follows:

[0016] Prior to testing, a reference nonlinear factor is calculated based on the second ultrasonic response signal acquired on the calibration block.

[0017] The feature extraction module is further configured to:

[0018] Before generating the nonlinear resonance spectrum distortion factor, the nonlinear resonance spectrum distortion factor is compared with the reference nonlinear factor;

[0019] If the comparison result is less than the first preset threshold, the nonlinear resonance spectrum distortion factor is input into the first fatigue damage physical model.

[0020] According to the technical solution provided in this application, the feature extraction module is further configured as follows:

[0021] If the comparison result is greater than or equal to the first preset threshold, a mode switching instruction is generated;

[0022] The device also includes a multimodal diagnostic module that is activated in response to the mode switching command;

[0023] The multimodal diagnostic module includes:

[0024] The pulsed ultrasonic excitation unit is used to control the ultrasonic excitation and acquisition module to re-probe the aluminum alloy weld to be tested with multiple preset discrete single-frequency high-energy pulse signals and acquire the corresponding third ultrasonic response signal.

[0025] The laser micro-vibration velocimetry unit is integrated on the side of the high-frequency piezoelectric probe and is used to synchronously collect the micro-vibration response of the weld surface when the pulsed ultrasonic excitation unit is working.

[0026] The data fusion unit is used to extract pulse nonlinear parameters based on the third ultrasonic response signal, extract vibration damping characteristics based on the micro vibration response, and fuse the pulse nonlinear parameters and the vibration damping characteristics to generate a fusion damage index.

[0027] The life assessment module is configured to receive the fusion damage index to replace the nonlinear resonance spectrum distortion factor after the multimodal diagnostic module is started, and to calculate the remaining fatigue life.

[0028] According to the technical solution provided in this application, the life assessment module further includes:

[0029] The second fatigue damage physical model takes the fusion damage index as input and outputs the second remaining fatigue life.

[0030] The model selector is configured to switch between the first fatigue damage physical model and the second fatigue damage physical model based on the type of received input parameters.

[0031] According to the technical solution provided in this application, the first fatigue damage physical model is a parameterized analytical model based on continuous damage mechanics, and the second fatigue damage physical model is a deep neural network model trained with multi-physics coupling data.

[0032] Secondly, this application proposes a rapid detection method for evaluating the fatigue life of aluminum alloy welds, implemented based on the rapid detection device for evaluating the fatigue life of aluminum alloy welds as described above, comprising the following steps:

[0033] The high-frequency piezoelectric probe is coupled to the calibration block to perform ultrasonic excitation and signal acquisition, obtain the second ultrasonic response signal, and calculate the reference nonlinear factor;

[0034] A high-energy, continuously sweeping ultrasonic excitation signal is injected into the aluminum alloy weld to be tested, the first ultrasonic response signal is acquired simultaneously, the higher-order harmonic components are demodulated and the relative energy ratio is calculated, and the nonlinear resonance spectrum distortion factor is generated.

[0035] Calculate the deviation between the nonlinear resonance spectrum distortion factor and the reference nonlinear factor;

[0036] If the deviation value is less than the first preset threshold, then enter the rapid evaluation mode; if the deviation value is greater than or equal to the first preset threshold, then enter the precise diagnosis mode.

[0037] The rapid assessment mode is executed by the first fatigue damage physical model, and the precise diagnosis mode is executed by the multimodal diagnosis module.

[0038] According to the technical solution provided in this application, a miniature semiconductor cooling / heating unit and a temperature sensor are integrated on the calibration test block to form an active temperature control calibration module;

[0039] Following the calculation of the baseline nonlinearity factor, the following steps are also included:

[0040] The active temperature control calibration module is controlled to maintain steady-state temperature at multiple different preset temperature points;

[0041] At each preset temperature point, ultrasonic excitation and signal acquisition are performed, and a preset nonlinear factor at that preset temperature point is calculated.

[0042] Based on each preset temperature point and its corresponding preset nonlinear factor, a temperature-nonlinear factor characteristic curve specifically for the current detection system and environment is generated in real time through curve fitting.

[0043] According to the technical solution provided in this application, the calculation of the benchmark nonlinear factor includes the following steps:

[0044] Get the current ambient temperature;

[0045] The temperature-nonlinear factor characteristic curve is retrieved to obtain the preset nonlinear factor corresponding to the current ambient temperature, and this is used as the reference nonlinear factor.

[0046] According to the technical solution provided in this application, after repeatedly inspecting the same aluminum alloy weld to be tested, the following steps are also included:

[0047] Store the remaining fatigue life results and their timestamps from each detection to form a time-series dataset of the weld's life decay.

[0048] The time series dataset of life decay is fitted to predict the target prediction time when the remaining life of the weld reaches the safety threshold.

[0049] Generate predictive maintenance information that includes the target prediction time.

[0050] According to the technical solution provided in this application, the method further includes the following steps:

[0051] The number and proportion of times the precision diagnostic mode was triggered in the historical inspection of the weld seam are automatically retrieved to obtain the data quality factor;

[0052] If the data quality factor indicates that the reliability of historical data is low, the next inspection cycle for the weld is shortened, and the multimodal diagnostic module is forcibly activated in the next inspection.

[0053] Compared with existing technologies, the advantages of this application are as follows: This application replaces the separate mode of two independent transducers with an integrated design of a high-frequency piezoelectric probe and receiver. This achieves stable and reliable detection results: the integrated design ensures that the injection of the ultrasonic excitation signal and the acquisition of the response signal modulated by the weld are completed synchronously at the same location and along the same path. This completely eliminates the positioning error and propagation path difference caused by the separation of the transmitting and receiving probes, significantly improving signal consistency and measurement repeatability. It also enhances the applicability and efficiency of detecting complex weld structures: the integrated probe is easier to arrange and couple on irregular surfaces such as curved and undulating welds, solving the pain points of difficult installation and inconsistent coupling states of separate probes in the weld area, making rapid, on-site detection of aluminum alloy welds possible. Attached Figure Description

[0054] Figure 1 A schematic diagram of the structure of a rapid testing device for evaluating the fatigue life of aluminum alloy welds provided in an embodiment of this application;

[0055] Figure 2 A flowchart illustrating the steps of a rapid detection method for evaluating the fatigue life of aluminum alloy welds, provided in an embodiment of this application.

[0056] The text labels in the image represent:

[0057] 1. Feature extraction module; 2. Life assessment module; 3. Ultrasonic excitation and acquisition module. Detailed Implementation

[0058] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0059] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0060] Example 1

[0061] As mentioned in the background section, in view of the problems in the prior art, this application proposes a rapid detection device for evaluating the fatigue life of aluminum alloy welds, such as... Figure 1 As shown, it includes:

[0062] Ultrasonic excitation and acquisition module 3 includes:

[0063] Broadband power amplifier;

[0064] A high-frequency piezoelectric probe, connected to the broadband power amplifier, is used to contact the aluminum alloy weld to be tested and is configured to inject a set of high-energy, continuously sweeping ultrasonic excitation signals into the weld.

[0065] The receiver is integrated with the high-frequency piezoelectric probe and is used to synchronously acquire the first ultrasonic response signal modulated by the micro-damage inside the weld.

[0066] Feature extraction module 1, electrically connected to the ultrasonic excitation and acquisition module 3, is configured as follows:

[0067] Demodulate the higher-order harmonic components caused by material fatigue damage from the first ultrasonic response signal;

[0068] Calculate the relative energy ratio between the higher-order harmonic components and the fundamental component to generate the nonlinear resonance spectrum distortion factor;

[0069] The life assessment module 2 has a first fatigue damage physical model embedded in it. The first fatigue damage physical model is configured to calculate and output the first remaining fatigue life of the aluminum alloy weld to be tested, using the nonlinear resonance spectrum distortion factor as the input parameter.

[0070] Specifically, the broadband power amplifier amplifies the low-power swept-frequency electrical signal generated by the signal generator. "Broadband" here refers to the amplifier having a sufficiently wide operating frequency range (e.g., 1MHz to 20MHz) to ensure distortion-free amplification of the continuously swept ultrasonic excitation signal. It provides the high-energy portion for driving the probe. The high-frequency piezoelectric probe is the component that directly contacts the aluminum alloy weld seam under test. "High-frequency" indicates that its operating frequency is typically in the megahertz range to meet the high resolution required for detecting microscopic damage. "Piezoelectric" refers to its core material being piezoelectric ceramic or single crystal, enabling the conversion between electrical and acoustic energy. This probe is configured to inject a set of high-energy, continuously swept-frequency ultrasonic excitation signals into the weld seam. Continuous frequency sweep means that the frequency of the excitation signal changes linearly or non-linearly from a starting frequency to a termination frequency within a preset time period (e.g., a few milliseconds to tens of milliseconds). This is fundamentally different from traditional single-frequency pulse train excitation; it can excite a wide-band response at once, like illuminating the material with a beam of acoustic white light, and can more comprehensively capture the non-linear behavior of the material at different frequencies. The receiver and the high-frequency piezoelectric probe are integrated as a single unit. This means that the receiving sensing element (which can be the same piezoelectric chip, switched to receiving mode instantly after transmission via an electronic switch; or it can be another dedicated receiving chip tightly integrated within the same probe housing) and the transmitting chip are physically inseparable. This design ensures that the injection of the excitation signal and the acquisition of the response signal are synchronously performed at exactly the same location and along the same acoustic propagation path, fundamentally eliminating errors caused by differences in installation position and coupling state in separate probes. Feature extraction module 1 is typically composed of a high-speed digital signal processor (DSP) or a field-programmable gate array (FPGA) equipped with a dedicated algorithm. This module receives the first ultrasonic response signal from the receiver (i.e., the raw voltage signal acquired from the weld). Demodulation of higher-order harmonic components: It first performs a Fast Fourier Transform (FFT) on the first ultrasonic response signal, converting the time-domain signal into a frequency-domain spectrum. In the frequency domain, in addition to the fundamental component (frequency f) corresponding to the excitation frequency, nonlinear elastic effects caused by material fatigue damage (such as dislocations, microplastic regions, and microcracks) also generate higher-order harmonic components (i.e., second and third harmonics, etc.) with frequencies of 2f, 3f, etc. The nonlinear resonance spectrum distortion factor is generated: this module calculates the ratio of the total energy of these higher-order harmonics (e.g., second and third harmonics can be selected) to the energy of the fundamental component, defining this ratio as the nonlinear resonance spectrum distortion factor. This factor is a dimensionless parameter that quantifies the severity of ultrasonic waveform distortion caused by material damage. The more severe the damage, the stronger the nonlinear effect, the higher the energy of the higher-order harmonics, and the larger the distortion factor. The life assessment module 2 is typically implemented by a microprocessor (such as a CPU). Its embedded "first fatigue damage physical model" is a mathematical model that establishes a quantitative relationship between the nonlinear resonance spectrum distortion factor and the remaining fatigue life.This module takes the calculated distortion factor as input, calculates through the model, and directly outputs a specific life index, namely the first remaining fatigue life, which can be a predicted value expressed in terms of cycle number or flight hours.

[0071] The technical principle is described below: When high-energy ultrasonic waves propagate in a material with fatigue damage, the damaged region (such as dislocation pinning-unpinning or contact nonlinearity of microcrack surfaces) acts like a nonlinear spring, causing the peak of the sine wave to propagate slightly faster than the trough, resulting in waveform distortion and the generation of harmonics. The more severe the fatigue damage, the more pronounced this nonlinear behavior. Therefore, by accurately measuring the energy of the harmonics relative to the fundamental wave (i.e., the distortion factor), the damage state of the material can be deduced, and its remaining life can be extrapolated using an established physical model of damage evolution.

[0072] In a preferred embodiment, the device further includes a calibration test block whose material and heat treatment state are the same as the initial non-service state of the aluminum alloy weld to be tested;

[0073] The feature extraction module 1 is configured as follows:

[0074] Prior to testing, a reference nonlinear factor is calculated based on the second ultrasonic response signal acquired on the calibration block.

[0075] The feature extraction module 1 is further configured to:

[0076] Before generating the nonlinear resonance spectrum distortion factor, the nonlinear resonance spectrum distortion factor is compared with the reference nonlinear factor;

[0077] If the comparison result is less than the first preset threshold, the nonlinear resonance spectrum distortion factor is input into the first fatigue damage physical model.

[0078] Specifically, the calibration block serves as a reference. Its material (e.g., aluminum alloy of the same grade) and heat treatment state (e.g., the same solution treatment and aging processes) must be identical to the initial unused state of the weld to be tested. This means that the block represents the baseline health state of the weld under test in terms of microstructure. It can be a cuboid or cylindrical block with a surface finish similar to the weld inspection area. The calibration function of feature extraction module 1 calculates the baseline nonlinearity factor: Before formal testing begins, the operator needs to couple the same high-frequency piezoelectric probe onto the calibration block. The device executes the exact same procedure as the formal testing: injecting a sweep frequency signal, acquiring a second ultrasonic response signal (from the healthy block), and then feature extraction module 1 calculates a baseline nonlinearity factor using the exact same algorithm. This factor characterizes the inherent background nonlinearity level of a healthy material under the current testing system conditions (including probe performance, amplifier gain, ambient temperature, etc.). Comparison and Threshold Judgment: After generating the nonlinear resonance spectrum distortion factor of the weld to be tested, the module does not directly input it into the life assessment model, but first compares it with the benchmark nonlinear factor. This comparison can be calculated by calculating the relative difference or ratio. The system presets a first threshold, which is determined through a large number of experiments, to distinguish between measurement noise and real damage signals. Only when the comparison result (e.g., the increment of the weld distortion factor relative to the benchmark value) is greater than or equal to this threshold is a meaningful damage signal considered to have been detected, and it is then input into the first fatigue damage physical model for calculation.

[0079] In a preferred embodiment, the feature extraction module 1 is further configured to:

[0080] If the comparison result is greater than or equal to the first preset threshold, a mode switching instruction is generated;

[0081] The device also includes a multimodal diagnostic module that is activated in response to the mode switching command;

[0082] The multimodal diagnostic module includes:

[0083] The pulsed ultrasonic excitation unit is used to control the ultrasonic excitation and acquisition module 3 to re-probe the aluminum alloy weld to be tested with multiple preset discrete single-frequency high-energy pulse signals and acquire the corresponding third ultrasonic response signal.

[0084] The laser micro-vibration velocimetry unit is integrated on the side of the high-frequency piezoelectric probe and is used to synchronously collect the micro-vibration response of the weld surface when the pulsed ultrasonic excitation unit is working.

[0085] The data fusion unit is used to extract pulse nonlinear parameters based on the third ultrasonic response signal, extract vibration damping characteristics based on the micro vibration response, and fuse the pulse nonlinear parameters and the vibration damping characteristics to generate a fusion damage index.

[0086] The life assessment module 2 is configured to receive the fused damage index to replace the nonlinear resonance spectrum distortion factor after the multimodal diagnostic module is started, and to calculate the remaining fatigue life.

[0087] Specifically, the pulsed ultrasonic excitation unit: This unit, controlled by software or hardware, takes over the command of the ultrasonic excitation and acquisition module 3. It commands the broadband power amplifier to stop sending continuous frequency sweep signals and instead generate a series of preset discrete single-frequency high-energy pulse signals. For example, three typical center frequency points (e.g., 5MHz, 10MHz, 15MHz) can be preset in the system, and then short-time pulse trains (Tone Burst signals) of each frequency can be emitted sequentially. Each pulse train contains a certain number of sine wave periods (e.g., 10 periods), with an envelope of a rectangular or Hanning window. The instantaneous energy of this signal is much higher than that of continuous waves, and it can more strongly excite the nonlinear response of the material. Data acquisition: For each emitted single-frequency pulse, the receiver synchronously acquires the corresponding third ultrasonic response signal. Because it is pulse excitation, this response signal is distinguishable in the time domain, facilitating independent analysis.

[0088] Specifically, the laser micro-vibration velocimetry unit is integrated as follows: This optical measurement unit is physically integrated into the side housing of the high-frequency piezoelectric probe, ensuring that the laser measurement point and the ultrasonic excitation / receiving area are spatially very close (e.g., less than 5 mm apart). This unit is triggered simultaneously with the pulsed ultrasonic excitation unit. It emits a focused laser beam towards the weld surface and receives its reflected light. Utilizing the principle of a laser Doppler vibrometer, the micro-vibration response of the weld surface is synchronously acquired by detecting the frequency shift (Doppler effect) of the reflected light caused by the micro-vibration of the material surface. This response is a time-domain signal, recording the instantaneous vibration velocity or displacement of the material surface caused by the ultrasonic wave propagation. This measurement is non-contact, avoiding the influence of added mass, and has extremely high frequency response and resolution.

[0089] Specifically, the data fusion unit extracts parameters for pulse nonlinear parameters. This unit performs spectral analysis on each group of acquired third-order ultrasonic response signals. Since the excitation is a single-frequency pulse, the second and third harmonic amplitudes generated in its response spectrum are very clear. This unit calculates the ratio of these harmonic amplitudes to the fundamental amplitude, or further calculates the second and third-order nonlinear coefficients, collectively referred to as pulse nonlinear parameters. This parameter has a similar physical meaning to the distortion factor, but due to its different excitation method, its numerical value and sensitivity characteristics also differ, providing another dimension of acoustic nonlinear information.

[0090] Specifically, regarding vibration damping characteristics: This unit analyzes the acquired time-domain signal of microscopic vibration response. It extracts the free decay segment of the vibration after the pulse and calculates its logarithmic decay rate or damping ratio by exponentially fitting the decay curve. Fatigue damage within the material (such as microcracks and increased dislocation density) increases internal friction, leading to faster dissipation of vibration energy, which manifests as increased damping characteristics.

[0091] Specifically, data fusion: This unit fuses parameters from two different physical sources—impulse nonlinearity parameters (characterizing material elastic nonlinearity) and vibration damping characteristics (characterizing material energy dissipation). The fusion algorithm can be a simple weighted summation, a more complex multi-parameter regression model, or a small neural network. Its output is a comprehensive fused damage index. The advantage of this index is that it simultaneously reflects the material's damage state in both acoustic and dynamic aspects, providing richer information dimensions.

[0092] Specifically, the input switching for life assessment: After the precise multimodal diagnostic mode is activated, life assessment module 2 will temporarily ignore the nonlinear resonance spectrum distortion factor initially obtained from the fast mode, and instead receive this more robust and comprehensive fused damage index as input to perform the final calculation of the remaining fatigue life.

[0093] In a preferred embodiment, the life assessment module 2 further includes:

[0094] The second fatigue damage physical model takes the fusion damage index as input and outputs the second remaining fatigue life.

[0095] The model selector is configured to switch between the first fatigue damage physical model and the second fatigue damage physical model based on the type of received input parameters.

[0096] Specifically, the second fatigue damage physical model is an independent life prediction model that runs parallel to the first fatigue damage physical model. Its input specifically refers to the fused damage index from the multimodal diagnostic module. This model is built upon multiphysics coupled data; that is, the data used during model training includes not only ultrasonic nonlinear parameters but also vibration damping data, and their correspondence with remaining life at different fatigue stages. Model selector: This is a logic control unit (which can be implemented by a software algorithm). It is configured to automatically switch between the first and second fatigue damage physical models based on the type of received input parameters. When the system is in fast evaluation mode, the feature extraction module 1 outputs the nonlinear resonance spectrum distortion factor. After recognizing this parameter, the model selector automatically selects and calls the first fatigue damage physical model for calculation. When the system enters precise diagnostic mode, the multimodal diagnostic module ultimately outputs the fused damage index. After recognizing this completely different parameter, the model selector automatically switches to the second fatigue damage physical model for calculation.

[0097] In a preferred embodiment, the first fatigue damage physical model is a parameterized analytical model based on continuous damage mechanics, and the second fatigue damage physical model is a deep neural network model trained with multi-physics coupling data.

[0098] Specifically, the first fatigue damage physical model is a parameterized analytical model based on continuum damage mechanics (CDM). This model is based on physical laws and has a clear mathematical expression. Continuum damage mechanics treats damage within a material (such as micropores and microcracks) as a continuously distributed field variable, characterized by the damage variable D. D=0 represents material integrity, and D=1 represents complete material failure. This model describes the evolution of the damage variable D with load cycles N through differential equations. A commonly used CDM model is the Chaboche-Lemaitre model, whose evolution equation is typically in the form: dD / dN = [σ_a / (M * (1- D))]^m * f(σ_m,...). Here, σ_a is the stress amplitude, M and m are material parameters, and f is a function of the mean stress σ_m, etc. Parameterization: In this application, the model input is no longer the traditional stress parameter, but rather the nonlinear resonance spectrum distortion factor (denoted as η). Therefore, it is necessary to establish a relationship between η and the driving force (such as effective stress) within the model. In practice, extensive fatigue tests on aluminum alloy welds are conducted to calibrate the material parameters (such as M and m) in the damage evolution equation and establish a quantitative relationship between η and the damage variable D, for example, through the empirical formula D = A * η. B(A and B are fitting parameters) Correlation. Ultimately, the model is implemented as a piece of program code or an analytical calculator embedded in the life assessment module 2. When η is input, the model first calculates the current damage D, and then predicts the remaining number of cycles required for D to reach a critical value (e.g., 0.7) based on the integral of the evolution equation, i.e., the first remaining fatigue life.

[0099] Specifically, the second fatigue damage physical model is a data-driven black-box or gray-box model. A deep neural network (DNN) is a machine learning model inspired by the neural networks of the human brain, consisting of multiple layers (input layer, hidden layers, output layer), each containing a large number of neurons. It makes predictions by learning complex nonlinear mappings from a large amount of known data. First, a dedicated training dataset needs to be constructed. This dataset comes from a large amount of experimental data on aluminum alloy weld specimens covering different damage states. For each specimen, its "multi-physics coupling data" needs to be collected, namely: pulse nonlinear parameters (from pulsed ultrasonic excitation), vibration damping characteristics (from laser micro-vibration velocimetry), and known true remaining fatigue life (obtained through interrupted fatigue testing). These data together constitute the input features (pulse nonlinear parameters and vibration damping characteristics) and output label (true remaining fatigue life) of the DNN. Network training: This dataset is used to train a DNN model (e.g., a fully connected network with 3 hidden layers and 128 neurons per layer). The training process is an iterative optimization process. Through the backpropagation algorithm, the network's tens of thousands of connection weights and biases are continuously adjusted to minimize the error (e.g., mean squared error) between the network's predicted output (predicted lifetime) and the actual output (actual lifetime) for the training samples. Model deployment: After training is completed and validated, the topology and all parameters (weights and biases) of this DNN model are fixed and embedded into lifetime assessment module 2. During actual detection, when the data fusion unit generates a fused damage index (i.e., a feature vector containing pulse nonlinear parameters and vibration damping characteristics), this vector is directly input into the trained DNN. The DNN calculates through forward propagation and directly outputs the "second remaining fatigue life".

[0100] Cooperative operation of model selectors:

[0101] Path 1 (Fast Evaluation): When the input is a single "nonlinear resonance spectrum distortion factor", the selector calls the CDM analytical model. This is the optimal choice because the CDM model has a simple structure, fast computation speed, and low resource consumption, making it very suitable for the efficiency requirements of rapid screening. Furthermore, the physics-based model can provide physically reliable extrapolations for such a single, explicit input parameter.

[0102] Path Two (Precise Diagnosis): When the input is a high-dimensional "fusion damage index," the selector calls the DNN model. This is the optimal choice because DNNs are inherently designed to handle high-dimensional, multimodal inputs. It can fully exploit the complex complementary and reinforcing relationships between pulse nonlinear parameters and vibration damping characteristics. The knowledge learned from massive amounts of data can achieve a mapping more accurate than any simple formula, thus providing the most accurate life prediction under the richest information conditions.

[0103] Example 2

[0104] Building upon Example 1, this example proposes a rapid detection method for evaluating the fatigue life of aluminum alloy welds, implemented using the rapid detection device for evaluating the fatigue life of aluminum alloy welds as described above. Figure 2 As shown, it includes the following steps:

[0105] S1. Couple the high-frequency piezoelectric probe to the calibration test block, perform ultrasonic excitation and signal acquisition, obtain the second ultrasonic response signal, and calculate the reference nonlinear factor;

[0106] S2. Inject a high-energy, continuously sweeping ultrasonic excitation signal into the aluminum alloy weld to be tested, simultaneously acquire the first ultrasonic response signal, demodulate the higher-order harmonic components and calculate the relative energy ratio, and generate the nonlinear resonance spectrum distortion factor.

[0107] S3. Calculate the deviation between the nonlinear resonance spectrum distortion factor and the reference nonlinear factor;

[0108] S4. If the deviation value is less than the first preset threshold, then enter the rapid evaluation mode; if the deviation value is greater than or equal to the first preset threshold, then enter the precise diagnosis mode.

[0109] The rapid assessment mode is executed by the first fatigue damage physical model, and the precise diagnosis mode is executed by the multimodal diagnosis module.

[0110] Furthermore, the calibration block integrates a miniature semiconductor cooling / heating unit and a temperature sensor to form an active temperature control calibration module;

[0111] Following the calculation of the baseline nonlinearity factor, the following steps are also included:

[0112] The active temperature control calibration module is controlled to maintain steady-state temperature at multiple different preset temperature points;

[0113] At each preset temperature point, ultrasonic excitation and signal acquisition are performed, and a preset nonlinear factor at that preset temperature point is calculated.

[0114] Based on each preset temperature point and its corresponding preset nonlinear factor, a temperature-nonlinear factor characteristic curve specifically for the current detection system and environment is generated in real time through curve fitting.

[0115] Specifically, the miniature semiconductor cooling / heating unit operates through the Peltier effect; when a direct current flows through it, one end cools while the other heats; reversing the current direction interchanges the hot and cold ends. This unit is tightly integrated into one side or bottom of the calibration block, typically bonded to it with highly thermally conductive silicone grease for efficient heat exchange. Its power and size are specially designed to allow for controlled temperature changes within minutes. The temperature sensor (such as a PT100 platinum resistance thermometer or DS18B20) is embedded or closely attached to the inside or surface of the calibration block to monitor its current temperature accurately and in real time.

[0116] Detailed implementation steps:

[0117] Step 1: Initial baseline calculation. First, an initial baseline nonlinearity factor is calculated at ambient temperature.

[0118] Step Two: Multi-Temperature Steady-State Excitation. The system controls the active temperature control calibration module to start working. It drives the semiconductor cooling / heating unit to sequentially set and stabilize the temperature of the calibration test block at multiple preset temperature points. For example, the preset temperature points may include 5℃, 15℃ (room temperature), 25℃, and 35℃ to cover the ambient temperature range where the equipment may operate. At each temperature point, after the temperature sensor reading stabilizes (e.g., maintained within ±0.5℃ of the target temperature for 1 minute), the system automatically executes an ultrasonic excitation and signal acquisition process, that is, controls the probe to inject a swept frequency signal into the calibration test block and acquires the response signal.

[0119] Step 3: Calculate the nonlinear factor at each temperature point. Feature extraction module 1 processes the response signal collected at each stable temperature point and calculates the preset nonlinear factor corresponding to that temperature point. At this point, the system obtains a set of data pairs: (T1, η1), (T2, η2), ..., (Tn, ηn), where T is the temperature and η is the corresponding nonlinear factor.

[0120] Step 4: Generate the temperature-nonlinear factor characteristic curve. The system's built-in algorithm (such as the least squares method) performs curve fitting on these data pairs. Since the material's nonlinear parameters typically have a nonlinear relationship with temperature, the fitting function may be a quadratic polynomial or an exponential function. After fitting, a "temperature-nonlinear factor characteristic curve" specific to this probe and calibration block combination is obtained. This curve is stored in the system as a temperature compensation benchmark for subsequent tests.

[0121] Further, the calculation of the baseline nonlinear factor includes the following steps:

[0122] Get the current ambient temperature;

[0123] The temperature-nonlinear factor characteristic curve is retrieved to obtain the preset nonlinear factor corresponding to the current ambient temperature, and this is used as the reference nonlinear factor.

[0124] Specifically, Step 1: The system obtains the current ambient temperature by reading the real-time readings of the temperature sensor integrated inside the probe or on the calibration block. This temperature is considered to be very close to the ambient temperature of the weld to be tested. Step 2: Retrieve the characteristic curve and perform interpolation calculation. The system retrieves the fitted temperature-nonlinear factor characteristic curve from memory. Then, using the current ambient temperature as input, it performs a lookup or interpolation calculation on the curve. If the current temperature is exactly at a preset temperature point, the preset nonlinear factor corresponding to that point is directly read. If the current temperature is between two preset temperature points (which is more common), algorithms such as linear interpolation are used to calculate the accurate preset nonlinear factor corresponding to the current temperature. Step 3: Set the baseline nonlinear factor. The preset nonlinear factor calculated or retrieved in Step 2 is directly used as the baseline nonlinear factor required for the current test.

[0125] Furthermore, after repeatedly inspecting the same aluminum alloy weld to be tested, the following steps are also included:

[0126] Store the remaining fatigue life results and their timestamps from each detection to form a time-series dataset of the weld's life decay.

[0127] The time series dataset of life decay is fitted to predict the target prediction time when the remaining life of the weld reaches the safety threshold.

[0128] Generate predictive maintenance information that includes the target prediction time.

[0129] Specifically, after each inspection of the same weld (whether in fast or precision mode), the system associates the inspection result (remaining fatigue life result, e.g., 125,000 cycles) with the timestamp of that inspection (e.g., 2023-10-27 14:30:01) and appends it to a database or file dedicated to that weld. After multiple inspections, a time-sorted sequence is formed, known as the "life decay time series dataset." For example: [(t1, L1),(t2, L2), ..., (tn, Ln)], where L represents the remaining life. The system periodically or upon user instruction performs data analysis on this time series dataset. It uses curve fitting algorithms (e.g., exponential decay fitting, power-law decay fitting, or damage evolution curve fitting based on a physical model) to find the mathematical function L(t) that best reflects the trend of the remaining life of the weld over time. Then, the system sets a "safety threshold" L_critical, which is pre-defined according to engineering safety specifications and represents the minimum remaining lifetime (e.g., 20,000 cycles). Subsequently, it solves the equation L(t) = L_critical to calculate the future time point at which the remaining lifetime reaches this threshold, i.e., the "target prediction time" T_predictive. The system automatically generates a predictive maintenance message, which includes at least the "target prediction time" T_predictive, and can optionally include the current remaining lifetime, decay rate, etc.

[0130] Furthermore, the method also includes the following steps:

[0131] The number and proportion of times the precision diagnostic mode was triggered in the historical inspection of the weld seam are automatically retrieved to obtain the data quality factor;

[0132] If the data quality factor indicates that the reliability of historical data is low, the next inspection cycle for the weld is shortened, and the multimodal diagnostic module is forcibly activated in the next inspection.

[0133] Specifically, the system automatically retrieves the complete inspection history of the weld. It counts the total number of times the precision diagnostic mode was triggered in each inspection. Then, it calculates the proportion of this number to the total number of inspections. This proportion is defined as the data quality factor Q, i.e., Q = (N_precise / N_total). A high Q value (e.g., Q>0.5) means that more than half of the inspections historically required a more complex confirmation process due to unreliable rapid screening results. This suggests that the damage signal of the weld may be weak and variable, or that the coupling conditions at the inspection location are not ideal. The system compares the calculated Q value with an internally preset threshold.

[0134] Decision 1: Shorten the inspection cycle. If the Q value indicates low reliability of historical data (e.g., Q is above a certain threshold Q_threshold), the system will automatically shorten the next inspection cycle for that weld. For example, if the original plan was to inspect again in 12 months, it will now be adjusted to 6 months. This is a risk control strategy that compensates for the lack of reliability of a single inspection by increasing the inspection frequency, and by more intensively tracking the evolution of potentially unstable damage.

[0135] Decision Two: Force the activation of precision diagnostics. Simultaneously, in the next test, regardless of the results of the rapid screening mode (even if the comparison result is less than the first preset threshold), the system will force the activation of the multimodal diagnostic module. This ensures that, even with historically poor data quality, the highest quality and most reliable test data can be obtained directly, skipping the rapid screening step that might again produce unreliable results.

[0136] This implementation achieves system self-learning and self-adaptation: the system is no longer a static execution tool, but can learn from historical performance and proactively adjust its behavior to adapt to the characteristics of different components, demonstrating a high degree of intelligence. Dynamic risk management: By monitoring data quality, high-risk components (i.e., those that are always difficult to assess) are subject to focused monitoring, optimizing the allocation of limited detection resources. Ensuring data reliability: The decision to forcibly activate the precision mode sacrifices some efficiency, but maximizes the accuracy of the final data in critical situations, reflecting the design principle of safety first.

[0137] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. The above descriptions are only preferred embodiments of this application. It should be noted that due to the limitations of written expression, while there are objectively infinite specific structures, those skilled in the art can make several improvements, modifications, or changes without departing from the principles of this invention, and can also combine the above technical features in an appropriate manner. These improvements, modifications, changes, or combinations, or the direct application of the inventive concept and technical solution to other situations without modification, should all be considered within the scope of protection of this application.

Claims

1. A rapid testing device for evaluating the fatigue life of aluminum alloy welds, characterized in that, include: The ultrasonic excitation and acquisition module (3) includes: Broadband power amplifier; A high-frequency piezoelectric probe, connected to the broadband power amplifier, is used to contact the aluminum alloy weld to be tested and is configured to inject a set of high-energy, continuously sweeping ultrasonic excitation signals into the weld. The receiver is integrated with the high-frequency piezoelectric probe and is used to synchronously acquire the first ultrasonic response signal modulated by the micro-damage inside the weld. The feature extraction module (1), electrically connected to the ultrasonic excitation and acquisition module (3), is configured as follows: Demodulate the higher-order harmonic components caused by material fatigue damage from the first ultrasonic response signal; Calculate the relative energy ratio between the higher-order harmonic components and the fundamental component to generate the nonlinear resonance spectrum distortion factor; The life assessment module (2) has a first fatigue damage physical model embedded in it. The first fatigue damage physical model is configured to calculate and output the first remaining fatigue life of the aluminum alloy weld to be tested using the nonlinear resonance spectrum distortion factor as the input parameter. The device also includes a calibration test block, the material and heat treatment of which are the same as the initial non-service state of the aluminum alloy weld to be tested; The feature extraction module (1) is configured as follows: Prior to testing, a reference nonlinear factor is calculated based on the second ultrasonic response signal acquired on the calibration block. The feature extraction module (1) is further configured as follows: Before generating the nonlinear resonance spectrum distortion factor, the nonlinear resonance spectrum distortion factor is compared with the reference nonlinear factor; If the comparison result is less than the first preset threshold, the nonlinear resonance spectrum distortion factor is input into the first fatigue damage physical model. The feature extraction module (1) is further configured as follows: If the comparison result is greater than or equal to the first preset threshold, a mode switching instruction is generated; The device also includes a multimodal diagnostic module that is activated in response to the mode switching command; The multimodal diagnostic module includes: The pulsed ultrasonic excitation unit is used to control the ultrasonic excitation and acquisition module (3) to re-probe the aluminum alloy weld to be tested with multiple preset discrete single-frequency high-energy pulse signals and acquire the corresponding third ultrasonic response signal. The laser micro-vibration velocimetry unit is integrated on the side of the high-frequency piezoelectric probe and is used to synchronously collect the micro-vibration response of the weld surface when the pulsed ultrasonic excitation unit is working. The data fusion unit is used to extract pulse nonlinear parameters based on the third ultrasonic response signal, extract vibration damping characteristics based on the micro vibration response, and fuse the pulse nonlinear parameters and the vibration damping characteristics to generate a fusion damage index. The life assessment module (2) is configured to receive the fusion damage index to replace the nonlinear resonance spectrum distortion factor after the multimodal diagnostic module is started, and to calculate the remaining fatigue life.

2. The rapid testing device for evaluating the fatigue life of aluminum alloy welds according to claim 1, characterized in that: The life assessment module (2) also includes: The second fatigue damage physical model takes the fusion damage index as input and outputs the second remaining fatigue life. The model selector is configured to switch between the first fatigue damage physical model and the second fatigue damage physical model based on the type of received input parameters.

3. The rapid testing device for evaluating the fatigue life of aluminum alloy welds according to claim 2, characterized in that: The first fatigue damage physical model is a parameterized analytical model based on continuous damage mechanics, and the second fatigue damage physical model is a deep neural network model trained with multi-physics coupling data.

4. A rapid detection method for evaluating the fatigue life of aluminum alloy welds, implemented based on the rapid detection device for evaluating the fatigue life of aluminum alloy welds as described in any one of claims 1-3, characterized in that: Includes the following steps: The high-frequency piezoelectric probe is coupled to the calibration block to perform ultrasonic excitation and signal acquisition, obtain the second ultrasonic response signal, and calculate the reference nonlinear factor; A high-energy, continuously sweeping ultrasonic excitation signal is injected into the aluminum alloy weld to be tested, the first ultrasonic response signal is acquired simultaneously, the higher-order harmonic components are demodulated and the relative energy ratio is calculated, and the nonlinear resonance spectrum distortion factor is generated. Calculate the deviation between the nonlinear resonance spectrum distortion factor and the reference nonlinear factor; If the deviation value is less than the first preset threshold, then enter the rapid evaluation mode; if the deviation value is greater than or equal to the first preset threshold, then enter the precise diagnosis mode. The rapid assessment mode is executed by the first fatigue damage physical model, and the precise diagnosis mode is executed by the multimodal diagnosis module.

5. The rapid detection method for evaluating the fatigue life of aluminum alloy welds according to claim 4, characterized in that: The calibration block integrates a miniature semiconductor cooling / heating unit and a temperature sensor to form an active temperature control calibration module. Following the calculation of the baseline nonlinearity factor, the following steps are also included: The active temperature control calibration module is controlled to maintain steady-state temperature at multiple different preset temperature points; At each preset temperature point, ultrasonic excitation and signal acquisition are performed, and a preset nonlinear factor at that preset temperature point is calculated. Based on each preset temperature point and its corresponding preset nonlinear factor, a temperature-nonlinear factor characteristic curve specifically for the current detection system and environment is generated in real time through curve fitting.

6. The rapid detection method for evaluating the fatigue life of aluminum alloy welds according to claim 5, characterized in that: The calculation of the baseline nonlinear factor includes the following steps: Get the current ambient temperature; The temperature-nonlinear factor characteristic curve is retrieved to obtain the preset nonlinear factor corresponding to the current ambient temperature, and this is used as the reference nonlinear factor.

7. The rapid detection method for evaluating the fatigue life of aluminum alloy welds according to claim 4, characterized in that: After repeatedly inspecting the same aluminum alloy weld seam, the following steps are also included: Store the remaining fatigue life results and their timestamps from each detection to form a time-series dataset of the weld's life decay. The time series dataset of life decay is fitted to predict the target prediction time when the remaining life of the weld reaches the safety threshold. Generate predictive maintenance information that includes the target prediction time.

8. The rapid detection method for evaluating the fatigue life of aluminum alloy welds according to claim 4, characterized in that: The method also includes the following steps: The number and proportion of times the precision diagnostic mode was triggered in the historical inspection of the weld seam are automatically retrieved to obtain the data quality factor; If the data quality factor indicates that the reliability of historical data is low, the next inspection cycle for the weld is shortened, and the multimodal diagnostic module is forcibly activated in the next inspection.

Images