Crane counterbeam crack defect detection method and system

Abstract

The present invention provides a method and system for detecting crack defects in a balance beam of a crane, and relates to the technical field of defect detection. The method and system include arranging a plurality of ultrasonic guided wave sensor arrays on the surface of a balance beam of a crane to be detected, collecting ultrasonic guided wave signals inside the balance beam, and constructing a time-frequency domain feature vector of the ultrasonic guided wave of the balance beam; constructing a three-dimensional finite element simulation model of the balance beam of the crane, obtaining multiple groups of strain field data of crack defect expansion of the balance beam, and constructing a crack defect deformation ultrasonic-strain field fusion feature vector; using the crack defect deformation ultrasonic-strain field fusion feature vector to train a crack defect deformation detection intelligent discrimination model, obtaining the time-frequency domain feature vector and the strain distribution feature vector of the balance beam of the crane to be detected, inputting the trained crack defect deformation detection intelligent discrimination model into intelligent discrimination screening, and outputting the crack defect detection result of the balance beam to be detected, including the type, size, shape and specific position of the crack defect.

Description

Technical Field

[0001] The invention relates to defect detection technology, and in particular to a method and system for detecting crack defects in a crane balance beam. Background Art

[0002] Cranes are indispensable and important equipment in modern industry and construction. The balance beam is a key structural component that carries heavy objects, and its integrity directly affects the safety performance of the crane. During long-term use, the balance beam is prone to crack defects. If not discovered and handled in time, it may lead to serious safety accidents. Therefore, it is of great significance to effectively detect crack defects in the balance beam of the crane.

[0003] At present, the commonly used methods for detecting crack defects in crane balance beams mainly include visual inspection, ultrasonic testing, magnetic particle testing, etc. These traditional methods generally have problems such as low detection accuracy, low degree of automation, and difficulty in achieving online real-time monitoring. With the development of artificial intelligence technology, some researchers have tried to apply machine learning methods to crack defect detection, but they still face the following challenges: the feature information obtained by a single detection method is limited, and it is difficult to fully reflect the various characteristics of crack defects; there is a lack of in-depth analysis of the crack propagation process, and it is difficult to accurately predict the development trend of cracks; the generalization ability of the intelligent detection model is insufficient, and it is difficult to adapt to the detection needs under different working conditions and environments.

[0004] Therefore, it is urgent to develop a crane balance beam crack defect detection method that can comprehensively utilize multi-source information, has high precision and strong adaptability, so as to improve the accuracy, reliability and practicality of detection. Summary of the invention

[0005] The invention provides a method and system for detecting crack defects in a crane balance beam, which can solve the problems in the prior art.

[0006] A first aspect of the present invention,

[0007] A method for detecting crack defects in a crane balance beam is provided, comprising:

[0008] Arrange multiple ultrasonic guided wave sensor arrays on the surface of the balance beam of the crane to be tested, collect ultrasonic guided wave signals inside the balance beam, obtain ultrasonic guided wave signal data of the balance beam, perform wavelet time-frequency analysis, obtain a time-frequency energy distribution diagram of the ultrasonic guided wave of the balance beam, extract multi-scale and multi-directional texture features of the time-frequency energy distribution diagram of the ultrasonic guided wave of the balance beam, and construct a time-frequency domain feature vector of the ultrasonic guided wave of the balance beam;

[0009] Construct a three-dimensional finite element simulation model of a crane balance beam, set crack defects of different sizes, shapes and positions, perform crack defect extension simulation analysis, obtain multiple sets of balance beam crack defect extension strain field data, extract the strain distribution morphological characteristics, strain gradient characteristics and strain statistical characteristics of the multiple sets of balance beam crack defect extension strain field data, construct a balance beam crack defect strain field feature vector, fuse the balance beam ultrasonic guided wave time-frequency domain feature vector and the balance beam crack defect strain field feature vector, and construct a crack defect deformation ultrasonic-strain field fusion feature vector;

[0010] The crack defect deformation ultrasound-strain field fusion feature vector is used to train a crack defect deformation detection intelligent discrimination model, and the time-frequency domain feature vector and strain distribution feature vector of the crane balance beam to be detected are obtained. The trained crack defect deformation detection intelligent discrimination model is input for intelligent discrimination screening, and the crack defect detection result of the balance beam to be detected is output, including the type, size, shape and specific position of the crack defect.

[0011] Preferably,

[0012] Multiple ultrasonic guided wave sensor arrays are arranged on the surface of the balance beam of the crane to be tested, and the ultrasonic guided wave signals inside the balance beam are collected to obtain the ultrasonic guided wave signal data of the balance beam and perform wavelet time-frequency analysis to obtain the ultrasonic guided wave time-frequency energy distribution diagram of the balance beam, including:

[0013] Add Gaussian white noise of different scales to the collected ultrasonic guided wave signal inside the balance beam and perform ensemble empirical mode decomposition to obtain a series of intrinsic mode function components, calculate the correlation coefficient between each intrinsic mode function component and the ultrasonic guided wave signal inside the balance beam, automatically identify the noise-dominated intrinsic mode function components according to a preset correlation coefficient threshold and remove them, retain the intrinsic mode function components reflecting the ultrasonic guided wave signal inside the balance beam and reconstruct them, and obtain the denoised ultrasonic guided wave signal inside the balance beam;

[0014] The Hilbert transform is performed on the ultrasonic guided wave signal inside the balance beam after noise reduction to obtain an analytical signal, the envelope of the instantaneous amplitude of the analytical signal is extracted, and the local maximum point of the instantaneous amplitude is determined by using the local peak detection method, and the time interval between two adjacent local maximum points is compared with the preset peak interval threshold, and the time interval between the two local maximum points is determined as the boundary interval of different guided wave modes according to the comparison result;

[0015] Adaptively dividing the ultrasonic guided wave signal inside the balance beam after noise reduction according to the boundary interval, extracting the corresponding ultrasonic guided wave modal signal segment in each boundary interval, and adaptively separating different ultrasonic guided wave modal signals;

[0016] Constructing a wavelet basis function library, wherein the wavelet basis function library includes a plurality of wavelet basis function families, each wavelet basis function family is composed of wavelet basis functions of different orders and scale factors, and simultaneously extracting statistical features and frequency domain features of each separated ultrasonic guided wave modal signal segment to construct ultrasonic guided wave signal features;

[0017] Calculating the correlation coefficient between each wavelet basis function in the wavelet basis function library and the ultrasonic guided wave signal characteristics at different scales to obtain a correlation coefficient matrix, constructing a wavelet basis function adaptability measurement model based on the correlation coefficient matrix, using the output of the wavelet basis function adaptability measurement model as the fitness function, and searching the wavelet basis function library for the best matching wavelet basis function using a genetic algorithm;

[0018] Based on the obtained best matching wavelet basis function, an adaptive wavelet time-frequency transform is performed on the ultrasonic guided wave signal characteristics, and finally a time-frequency energy distribution diagram of the balanced beam ultrasonic guided wave is obtained.

[0019] Preferably,

[0020] Based on the obtained best matching wavelet basis function, the characteristics of the ultrasonic guided wave signal are subjected to adaptive wavelet time-frequency transformation, and finally the time-frequency energy distribution diagram of the balanced beam ultrasonic guided wave is obtained, including:

[0021] The ultrasonic guided wave signal characteristics are subjected to wavelet transform using the best matching wavelet basis function to obtain a time-frequency representation matrix, the energy distribution discreteness of the time-frequency representation matrix at different scales is analyzed, and the scale parameter of the wavelet transform is adaptively adjusted according to the minimum energy distribution discreteness criterion to obtain the optimal scale parameter range;

[0022] Within the optimal scale parameter range, the optimal translation parameter is obtained by optimizing the translation parameter of the best matching wavelet basis function and minimizing the mean square error between the spectrum of the best matching wavelet basis function and the characteristic spectrum of the ultrasonic guided wave signal;

[0023] Based on the obtained best matching wavelet basis function, optimal scale parameter and optimal translation parameter, the characteristics of the ultrasonic guided wave signal are subjected to adaptive wavelet time-frequency transformation, and finally a time-frequency energy distribution diagram of the balanced beam ultrasonic guided wave is obtained;

[0024] Among them, the calculation formulas of the optimal scale parameter and the optimal translation parameter are as follows:

[0025] ;

[0026] Among them, a * represents the optimal scale parameter, b * represents the optimal translation parameter, a represents the scale parameter, b represents the translation parameter, E(·) represents the energy distribution discreteness, μ represents the weight coefficient of the spectrum matching degree, S x(·) represents the frequency spectrum of ultrasonic guided wave signal characteristics, S Ψ (·) represents the frequency spectrum of the wavelet basis function, and f represents the frequency.

[0027] Preferably,

[0028] A three-dimensional finite element simulation model of the crane balance beam was constructed, and crack defects of different sizes, shapes and positions were set to perform crack defect propagation simulation analysis. Multiple sets of balance beam crack defect propagation strain field data were obtained, including:

[0029] On the basis of the standard finite element shape function of the three-dimensional finite element simulation model of the crane balance beam, a level set function describing the crack surface and a crack tip singular function describing the stress and strain singularity at the crack tip are introduced. The level set function is a signed distance function used to describe the signed distance from any point in space to the crack surface. The crack tip singular function is constructed by the linear elastic fracture mechanics theory.

[0030] The finite difference method is used to discretize the time and space derivatives of the level set function. The extension and bifurcation of the crack surface are determined by iteratively updating the level set function value. According to the updated level set function value, the level set function value at each finite element mesh node is calculated by finite element interpolation to determine the position of the crack surface.

[0031] The relative position relationship between the crack surface and the finite element mesh is determined according to the position of the crack surface, and the crack surface enhancement degree of freedom is dynamically updated using the geometric enhancement criterion. For the unit completely cut by the crack surface, the crack surface enhancement degree of freedom is introduced at its node;

[0032] According to the size of the crack tip plastic zone, the crack tip singularity enhancement degree of freedom is dynamically updated using the plastic zone estimation criterion. The size of the crack tip plastic zone is estimated using fracture mechanics theory. For the finite element mesh contained in the crack tip plastic zone, the crack tip singularity enhancement degree of freedom is introduced at the node.

[0033] Based on the crack surface enhanced degrees of freedom and crack tip singularity enhanced degrees of freedom, the stiffness matrix and load vector of the three-dimensional finite element simulation model of the crane balance beam are locally modified to obtain the target crane balance beam finite element simulation model. The target crane balance beam finite element simulation model is used to carry out crack defect propagation simulation analysis and obtain multiple sets of strain field data of crack defect propagation of the crane balance beam.

[0034] Preferably,

[0035] The finite element simulation model of the target crane balance beam is used to perform crack defect propagation simulation analysis, and multiple sets of strain field data of the crane balance beam crack defect propagation are obtained, including:

[0036] An initial crack defect is introduced at a preset position in the finite element simulation model of the target crane balance beam, and the signed distance value at the finite element mesh node containing the initial crack defect is calculated using the level set function. The crack surface enhancement degree of freedom is introduced into the finite element mesh node whose signed distance value is less than the preset distance threshold, so as to form a crack surface enhancement unit node set; and according to the size of the crack tip plastic zone, the crack tip singularity enhancement degree of freedom is introduced into the finite element mesh node located in the crack tip plastic zone, so as to form a crack tip singularity enhancement unit node set;

[0037] Apply loads and boundary conditions under actual working conditions to the finite element simulation model of the target crane balance beam, and select appropriate time steps and convergence criteria according to the nature of the load and the time scale of crack growth to perform crack growth analysis;

[0038] In the crack growth analysis process, for each time step, the strain field and stress field are calculated based on the node displacement field obtained in the previous time step, and the crack tip is judged whether it meets the growth condition according to the material fracture criterion. If the growth condition is met, the crack growth direction and crack growth increment are determined according to the maximum stress intensity factor principle;

[0039] The level set function value is updated according to the crack extension increment, and the signed distance value at the finite element mesh node is recalculated. For the new crack surface position, the crack surface enhancement unit node set and the crack tip singularity enhancement unit node set are updated. The iteration is repeated until the preset simulation time is reached, and finally multiple sets of strain field data of the crack defect extension of the crane balance beam are output.

[0040] Preferably,

[0041] Using the crack defect deformation ultrasound-strain field fusion feature vector to train the crack defect deformation detection intelligent discrimination model includes:

[0042] Based on the ultrasonic-strain field fusion characteristics of the crack defect deformation of the balance beam, the source domain dataset and the target domain dataset are constructed respectively, where the source domain dataset consists of labeled training samples, and the target domain dataset consists of unlabeled test samples;

[0043] Using the source domain dataset, the initial intelligent discrimination model is trained by minimizing the classification loss function on the source domain. The domain adaptation mechanism is introduced into the initial intelligent discrimination model. The conditional domain adversarial neural network method is used to achieve feature alignment between the source domain and the target domain by minimizing the domain discrimination loss function, thus obtaining a domain adaptation intelligent discrimination model.

[0044] The domain adaptation intelligent discrimination model is end-to-end trained by using source domain datasets and target domain datasets, and the domain adaptation loss function of the domain adaptation intelligent discrimination model is constructed, which includes classification loss and adversarial loss. The model parameters are updated by using an alternating optimization strategy, and the trained crack defect deformation detection intelligent discrimination model is finally obtained by minimizing the domain adaptation loss function.

[0045] Preferably,

[0046] The calculation formula of the domain adaptation loss function for constructing the domain adaptation intelligent discrimination model is as follows:

[0047] ;

[0048] Where L represents the domain adaptation loss function, n s represents the number of training samples in the source domain dataset, l (·) represents the classification loss function, F θ (·) represents the domain adaptation intelligent discriminant model, θ represents the parameters of the domain adaptation intelligent discriminant model, represents the i-th training sample in the source domain dataset, represents the true label of the i-th training sample in the source domain dataset, λ represents the hyperparameter for balancing classification loss and adversarial loss, and k δ (·) represents the domain discriminator, δ represents the domain discriminator parameter, n t represents the number of test samples in the target domain dataset, represents the jth test sample in the target domain dataset.

[0049] The second aspect of the present invention,

[0050] A crane balance beam crack defect detection system is provided, comprising:

[0051] The first unit is used to arrange multiple ultrasonic guided wave sensor arrays on the surface of the balance beam of the crane to be tested, collect ultrasonic guided wave signals inside the balance beam, obtain ultrasonic guided wave signal data of the balance beam and perform wavelet time-frequency analysis to obtain a time-frequency energy distribution diagram of the ultrasonic guided wave of the balance beam, extract multi-scale and multi-directional texture features of the time-frequency energy distribution diagram of the ultrasonic guided wave of the balance beam, and construct a time-frequency domain feature vector of the ultrasonic guided wave of the balance beam;

[0052] The second unit is used to construct a three-dimensional finite element simulation model of a crane balance beam, set crack defects of different sizes, shapes and positions, perform crack defect extension simulation analysis, obtain multiple sets of balance beam crack defect extension strain field data, extract the strain distribution morphological characteristics, strain gradient characteristics and strain statistical characteristics of the multiple sets of balance beam crack defect extension strain field data, construct a balance beam crack defect strain field feature vector, fuse the balance beam ultrasonic guided wave time-frequency domain feature vector and the balance beam crack defect strain field feature vector, and construct a crack defect deformation ultrasonic-strain field fusion feature vector;

[0053] The third unit is used to train a crack defect deformation detection intelligent discrimination model using the crack defect deformation ultrasound-strain field fusion feature vector, obtain the time-frequency domain feature vector and strain distribution feature vector of the crane balance beam to be detected, input the trained crack defect deformation detection intelligent discrimination model for intelligent discrimination screening, and output the crack defect detection result of the balance beam to be detected, including the type, size, shape and specific position of the crack defect.

[0054] The third aspect of the present invention,

[0055] An electronic device is provided, comprising:

[0056] processor;

[0057] a memory for storing processor-executable instructions;

[0058] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0059] A fourth aspect of the present invention,

[0060] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0061] Through the fusion of ultrasonic guided waves and strain field data, this method can detect smaller and more hidden crack defects and improve the sensitivity of detection. Combined with time-frequency analysis and multi-scale and multi-directional texture feature extraction, various characteristics of crack defects can be fully captured, including position, shape, size, etc. The use of machine learning models for intelligent discrimination reduces manual intervention and improves detection efficiency and accuracy. Through simulation analysis of crack propagation under different conditions, this method has strong adaptability and can cope with various complex crack situations. It can detect potential crack hazards early, not only to detect whether cracks exist, but also to give specific parameters of cracks, such as type, size, shape and location, which is conducive to preventive maintenance of equipment, prolong service life, reduce unnecessary downtime and maintenance time, reduce maintenance costs, and improve economic benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 It is a schematic flow chart of a method for detecting crack defects in a crane balance beam according to an embodiment of the present invention;

[0063] Figure 2 It is a structural schematic diagram of a crane balance beam crack defect detection system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0064] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0065] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0066] Figure 1 FIG. 1 is a flow chart of a method for detecting crack defects in a crane balance beam according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0067] S101. Arrange multiple ultrasonic guided wave sensor arrays on the surface of the balance beam of the crane to be tested, collect ultrasonic guided wave signals inside the balance beam, obtain ultrasonic guided wave signal data of the balance beam and perform wavelet time-frequency analysis to obtain a time-frequency energy distribution diagram of the ultrasonic guided wave of the balance beam, extract multi-scale and multi-directional texture features of the time-frequency energy distribution diagram of the ultrasonic guided wave of the balance beam, and construct a time-frequency domain feature vector of the ultrasonic guided wave of the balance beam.

[0068] Among them, ultrasonic guided waves are a type of stress wave that propagates in plate-like or shell-like structures. Due to their long propagation distance in the structure and their strong sensitivity to defects, they are widely used in the field of structural health monitoring and defect non-destructive testing. When ultrasonic guided waves propagate in a structure containing crack defects, stress concentration will occur at the crack tip, resulting in scattering and diffraction of ultrasonic guided waves. By analyzing the crack information carried by scattered waves and diffracted waves, the crack defects can be located and evaluated. In order to obtain comprehensive ultrasonic guided wave detection information of the balance beam, PZT piezoelectric ceramics are used as ultrasonic guided wave excitation and receiving transducers, which are pasted on the surface of the balance beam. Among them, the number of array elements is 8, and the circular ring is evenly distributed to achieve omnidirectional transmission and reception of ultrasonic guided wave signals.

[0069] In view of the non-stationary characteristics of ultrasonic guided wave signals, the wavelet time-frequency analysis method is introduced to simultaneously obtain the time and frequency information of the signal and more accurately characterize the propagation characteristics of ultrasonic guided waves. Optionally, the Morlet complex wavelet is taken as the basis function, and the signal is subjected to wavelet time-frequency transformation to obtain the wavelet coefficients under different scales and translations, and then the time-frequency energy distribution diagram of the ultrasonic guided wave signal is obtained.

[0070] In order to quantitatively describe the crack defect information contained in the ultrasonic guided wave time-frequency energy distribution map, it is necessary to extract sensitive feature parameters from the time-frequency image. Texture features are statistical features that describe the grayscale distribution law of an image and can reflect the grayscale change trend of an image. The grayscale co-occurrence matrix method is used to extract the texture features of the time-frequency distribution map from multiple scales and directions, including energy, grayscale average, grayscale variance, correlation, and contrast. Through multi-scale and multi-directional texture feature extraction, a 50-dimensional ultrasonic guided wave time-frequency domain feature vector is finally constructed.

[0071] In an optional embodiment,

[0072] Multiple ultrasonic guided wave sensor arrays are arranged on the surface of the balance beam of the crane to be tested, and the ultrasonic guided wave signals inside the balance beam are collected to obtain the ultrasonic guided wave signal data of the balance beam and perform wavelet time-frequency analysis to obtain the ultrasonic guided wave time-frequency energy distribution diagram of the balance beam, including:

[0073] Add Gaussian white noise of different scales to the collected ultrasonic guided wave signal inside the balance beam and perform ensemble empirical mode decomposition to obtain a series of intrinsic mode function components, calculate the correlation coefficient between each intrinsic mode function component and the ultrasonic guided wave signal inside the balance beam, automatically identify the noise-dominated intrinsic mode function components according to a preset correlation coefficient threshold and remove them, retain the intrinsic mode function components reflecting the ultrasonic guided wave signal inside the balance beam and reconstruct them, and obtain the denoised ultrasonic guided wave signal inside the balance beam;

[0074] The Hilbert transform is performed on the ultrasonic guided wave signal inside the balance beam after noise reduction to obtain an analytical signal, the envelope of the instantaneous amplitude of the analytical signal is extracted, and the local maximum point of the instantaneous amplitude is determined by using the local peak detection method, and the time interval between two adjacent local maximum points is compared with the preset peak interval threshold, and the time interval between the two local maximum points is determined as the boundary interval of different guided wave modes according to the comparison result;

[0075] Adaptively dividing the ultrasonic guided wave signal inside the balance beam after noise reduction according to the boundary interval, extracting the corresponding ultrasonic guided wave modal signal segment in each boundary interval, and adaptively separating different ultrasonic guided wave modal signals;

[0076] Constructing a wavelet basis function library, wherein the wavelet basis function library includes a plurality of wavelet basis function families, each wavelet basis function family is composed of wavelet basis functions of different orders and scale factors, and simultaneously extracting statistical features and frequency domain features of each separated ultrasonic guided wave modal signal segment to construct ultrasonic guided wave signal features;

[0077] Calculating the correlation coefficient between each wavelet basis function in the wavelet basis function library and the ultrasonic guided wave signal characteristics at different scales to obtain a correlation coefficient matrix, constructing a wavelet basis function adaptability measurement model based on the correlation coefficient matrix, using the output of the wavelet basis function adaptability measurement model as the fitness function, and searching the wavelet basis function library for the best matching wavelet basis function using a genetic algorithm;

[0078] Based on the obtained best matching wavelet basis function, an adaptive wavelet time-frequency transform is performed on the ultrasonic guided wave signal characteristics, and finally a time-frequency energy distribution diagram of the balanced beam ultrasonic guided wave is obtained.

[0079] Exemplarily, first, Gaussian white noise of different scales is added to the collected ultrasonic guided wave signal inside the balance beam and ensemble empirical mode decomposition is performed. Ensemble empirical mode decomposition is an adaptive signal decomposition method, which obtains a series of intrinsic mode function components by repeating empirical mode decomposition. The intrinsic mode function components represent the oscillation mode of the signal on different time scales. By adding Gaussian white noise of different scales to the original signal, multiple groups of ensemble empirical mode decomposition results can be obtained. Gaussian white noise refers to random noise that obeys Gaussian distribution and has a constant power spectral density, and scale refers to the standard deviation of the noise.

[0080] The correlation coefficient between each intrinsic mode function component and the ultrasonic guided wave signal inside the balance beam is calculated, and the noise-dominated intrinsic mode function components are automatically identified and eliminated according to the preset correlation coefficient threshold. The correlation coefficient reflects the degree of correlation between the two signals, and the value range is from negative one to one. The larger the absolute value of the correlation coefficient, the more correlated the two signals are. By calculating the correlation coefficient between the intrinsic mode function component and the original signal, it can be determined whether the component is dominated by noise. The components with correlation coefficients lower than the preset threshold are regarded as noise-dominated components and eliminated, and the components with correlation coefficients higher than the threshold are retained and the signal is reconstructed to obtain the ultrasonic guided wave signal inside the balance beam after noise reduction.

[0081] The Hilbert transform is performed on the ultrasonic guided wave signal inside the balance beam after noise reduction to obtain the analytical signal. The Hilbert transform is a linear transformation from time domain to time domain. The imaginary part of the analytical signal is obtained by shifting the conjugate phase of the original signal by 90 degrees, and then a complex analysis signal with physical meaning is constructed. The real part of the analytical signal is the original signal, and the imaginary part is the Hilbert transform of the original signal. The amplitude reflects the envelope characteristics of the original signal.

[0082] The instantaneous amplitude of the analytical signal is extracted by envelope, and the local maximum point of the instantaneous amplitude is determined by the local peak detection method. Envelope extraction refers to removing the high-frequency components in the instantaneous amplitude of the analytical signal to obtain an envelope curve that reflects the energy distribution of the signal. Local peak detection refers to finding the local maximum point in the envelope curve, that is, the point where the values ​​on both the left and right sides are smaller than the value of the point.

[0083] The time interval between two adjacent local maximum points is compared with a preset peak interval threshold, and the time interval between the two local maximum points is determined as the demarcation interval of different waveguide modes according to the comparison result. When ultrasonic guided waves propagate inside the balance beam, multiple waveguide modes will be excited. Different waveguide modes have different propagation speeds and dispersion characteristics, and the time intervals between their envelope peaks are also different. According to the demarcation interval, the ultrasonic guided wave signal inside the balance beam after noise reduction is adaptively divided, and the corresponding ultrasonic guided wave mode signal fragments are extracted in each demarcation interval to achieve adaptive separation of different ultrasonic guided wave mode signals. Since the demarcation intervals of different waveguide mode signals have been determined, the signal can be adaptively divided according to the position of the demarcation point, and the signal fragments corresponding to each waveguide mode are extracted to complete the adaptive separation of the waveguide mode signal.

[0084] Construct a wavelet basis function library, which contains multiple wavelet basis function families, and each wavelet basis function family is composed of wavelet basis functions of different orders and scale factors. Such as Morlet, Mexican hat, Daubechies, Symlets, Coiflets, etc., each type of wavelet basis function contains different orders and scale factors to form a complete set of wavelet basis functions. Wavelet transform is a time-scale signal analysis method that realizes multi-scale decomposition of signals by scaling and translating basis functions. Wavelet basis function is the basic analysis function in wavelet transform, and different wavelet basis functions have different time-frequency localization characteristics. The order of wavelet basis function reflects the smoothness of basis function, and the scale factor controls the time and frequency window width of basis function.

[0085] The statistical features and frequency domain features of each separated ultrasonic guided wave modal signal segment are extracted to construct the ultrasonic guided wave signal features. The statistical features reflect the time domain statistical properties of the signal. Common statistical features include mean, variance, peak, root mean square, etc. The frequency domain features reflect the frequency domain properties of the signal. Common frequency domain features include spectrum center, spectrum width, spectrum peak, etc.

[0086] The correlation coefficient between each wavelet basis function in the wavelet basis function library and the ultrasonic guided wave signal characteristics at different scales is calculated to obtain a correlation coefficient matrix. The correlation coefficient matrix reflects the degree of correlation between different wavelet basis functions and the ultrasonic guided wave signal characteristics at different scales. A wavelet basis function adaptability measurement model is constructed based on the correlation coefficient matrix, and the output of the wavelet basis function adaptability measurement model is used as the fitness function, and a genetic algorithm is used to search for the best matching wavelet basis function in the wavelet basis function library. The wavelet basis function adaptability measurement model quantitatively evaluates the degree of adaptation between the wavelet basis function and the signal to be analyzed by fusing multi-scale correlation coefficients. The genetic algorithm is a global optimization search algorithm that uses the fitness function as the optimization target and searches for the optimal solution through iterative evolution operations. Using the output of the wavelet basis function adaptability measurement model as the fitness function, a genetic algorithm can be used to quickly search for the wavelet basis function that best matches the ultrasonic guided wave signal in the wavelet basis function library. Finally, based on the obtained best matching wavelet basis function, an adaptive wavelet time-frequency transform is performed on the ultrasonic guided wave signal characteristics to obtain a balanced beam ultrasonic guided wave time-frequency energy distribution diagram.

[0087] In this embodiment, the adaptive wavelet time-frequency transform uses the wavelet basis function that best matches the signal to be analyzed to perform wavelet decomposition on the signal, which can effectively extract the local characteristics of the signal in the time-frequency domain and perform high-resolution representation of the time-frequency energy distribution of the signal. By performing adaptive wavelet time-frequency transform on the characteristics of the ultrasonic guided wave signal, a time-frequency energy distribution diagram reflecting the propagation law of the ultrasonic guided wave inside the balance beam and the defect sensitivity characteristics can be obtained, which provides an important basis for subsequent defect identification and quantitative evaluation.

[0088] In an optional embodiment,

[0089] Based on the obtained best matching wavelet basis function, the characteristics of the ultrasonic guided wave signal are subjected to adaptive wavelet time-frequency transformation, and finally the time-frequency energy distribution diagram of the balanced beam ultrasonic guided wave is obtained, including:

[0090] The ultrasonic guided wave signal characteristics are subjected to wavelet transform using the best matching wavelet basis function to obtain a time-frequency representation matrix, the energy distribution discreteness of the time-frequency representation matrix at different scales is analyzed, and the scale parameter of the wavelet transform is adaptively adjusted according to the minimum energy distribution discreteness criterion to obtain the optimal scale parameter range;

[0091] Within the optimal scale parameter range, the optimal translation parameter is obtained by optimizing the translation parameter of the best matching wavelet basis function and minimizing the mean square error between the spectrum of the best matching wavelet basis function and the characteristic spectrum of the ultrasonic guided wave signal;

[0092] Based on the obtained best matching wavelet basis function, optimal scale parameter and optimal translation parameter, the characteristics of the ultrasonic guided wave signal are subjected to adaptive wavelet time-frequency transformation, and finally a time-frequency energy distribution diagram of the balanced beam ultrasonic guided wave is obtained;

[0093] Among them, the calculation formulas of the optimal scale parameter and the optimal translation parameter are as follows:

[0094] ;

[0095] Among them, a * represents the optimal scale parameter, b * represents the optimal translation parameter, a represents the scale parameter, b represents the translation parameter, E(·) represents the energy distribution discreteness, μ represents the weight coefficient of the spectrum matching degree, S x (·) represents the frequency spectrum of ultrasonic guided wave signal characteristics, S Ψ (·) represents the frequency spectrum of the wavelet basis function, and f represents the frequency.

[0096] Exemplarily, the ultrasonic guided wave signal is subjected to wavelet transform using the best matching wavelet basis function. The wavelet transform algorithm can be implemented using a library function or self-written code to calculate a time-frequency representation matrix in which rows represent time and columns represent frequency.

[0097] For the time-frequency representation matrix, the energy distribution is calculated at different scales. Statistics such as the energy mean and variance at each scale can be calculated to measure the degree of energy dispersion. By combining parameters at different scales, the parameters that minimize the degree of energy distribution dispersion are selected.

[0098] Within the optimal scale parameter range, by optimizing the translation parameters of the wavelet basis function, the center frequency of the wavelet basis function in the frequency domain is aligned with the main frequency band of the ultrasonic guided wave signal, and the mean square error between the wavelet basis function spectrum and the signal spectrum is minimized to obtain the best translation parameters and improve the frequency resolution of wavelet time-frequency analysis. The optimized wavelet basis function and its scale and translation parameters are used to perform adaptive wavelet time-frequency transform on the ultrasonic guided wave signal, and the optimal time-frequency representation matrix that can accurately describe the time-frequency energy distribution characteristics of the ultrasonic guided wave signal is obtained, and finally the time-frequency energy distribution diagram of the balanced beam ultrasonic guided wave is obtained.

[0099] In this embodiment, by adaptively adjusting the scale and translation parameters of the wavelet transform, the method can better adapt to ultrasonic guided wave signals with different characteristics, thereby improving the flexibility and adaptability of signal analysis. By finding the best matching wavelet basis function and optimizing its parameters, a higher resolution can be obtained in both the time domain and the frequency domain, thereby more accurately capturing the time-frequency characteristics of the signal. By minimizing the energy distribution discreteness and the mean square error of the spectrum, the method can more accurately characterize the energy distribution of the ultrasonic guided wave signal, which is beneficial to subsequent feature extraction and analysis. The adaptive parameter optimization process helps to reduce the impact of noise on signal analysis and improves the signal processing capability in complex environments. More sophisticated time-frequency analysis helps to detect smaller signal changes, thereby improving the detection capability of small crack defects. The optimized time-frequency energy distribution diagram not only improves the accuracy and reliability of ultrasonic guided wave signal analysis, but also lays a more solid foundation for subsequent crack defect detection.

[0100] S102. Construct a three-dimensional finite element simulation model of a crane balance beam, set crack defects of different sizes, shapes and positions, perform crack defect expansion simulation analysis, obtain multiple sets of balance beam crack defect expansion strain field data, extract the strain distribution morphological characteristics, strain gradient characteristics and strain statistical characteristics of the multiple sets of balance beam crack defect expansion strain field data, construct a balance beam crack defect strain field feature vector, fuse the balance beam ultrasonic guided wave time-frequency domain feature vector and the balance beam crack defect strain field feature vector, and construct a crack defect deformation ultrasonic-strain field fusion feature vector.

[0101] For example, according to the actual structural parameters of the balance beam of the crane to be tested, a three-dimensional solid model of the balance beam is built in the finite element software ANSYS. On the basis of the three-dimensional model of the balance beam, crack defects are further introduced. Since the crack morphology and size are the key factors affecting the crack propagation, it is necessary to parametrically model the crack defects. The crack defect size parameters are defined as the crack length and depth, the morphological parameter is the aspect ratio, and semi-elliptical surface cracks of different sizes and shapes are implanted on the web in a parametric manner. To ensure the accuracy of the simulation, it is necessary to reasonably set the constitutive relationship and boundary conditions of the balance beam material. The balance beam of the crane generally adopts low-alloy high-strength structural steel, and its mechanical behavior can be described by a piecewise linear enhanced elastic-plastic constitutive model. The balance beam is hinged at both ends of the crane's luffing mechanism, and hinge constraints are applied to the hinged position in the finite element model. At the same time, under the maximum working load on the balance beam, a distributed stress load is applied to the web surface. By changing the crack size and morphological parameters, a series of crack defect extension conditions were set, and the elastic-plastic finite element analysis of crack extension was automatically performed using ANSYS to obtain the deformation and stress-strain responses of the balance beam at different crack extension stages.

[0102] In order to quantitatively describe the correspondence between crack defects and strain fields, it is necessary to extract sensitive features from the strain field. Specifically, the strain distribution morphological features, including the size of the strain concentration area, the center position of the strain concentration, the ellipticity of the strain distribution, etc., characterize the macroscopic strain distribution morphology caused by the crack; strain gradient features: calculate the gradient of the strain field, extract the mean value of the strain gradient amplitude, the consistency of the strain gradient direction, etc., to characterize the variation law of the strain gradient at the crack tip; strain statistical features: perform statistics on the strain field values, extract the strain mean, variance, skewness and kurtosis and other statistical quantities, and characterize the numerical distribution law of the strain field.

[0103] Crack defects of different sizes and shapes are implanted in the balance beam model by parametric modeling, and elastic-plastic finite element analysis of crack propagation is performed to obtain the strain field data caused by crack propagation. The morphological features, gradient features and statistical features are further extracted from the strain field to construct the crack defect strain field feature vector. This feature vector can quantitatively characterize the type, size, morphology and other parameters of the crack defect and reflect the crack propagation state.

[0104] In an optional embodiment,

[0105] A three-dimensional finite element simulation model of the crane balance beam was constructed, and crack defects of different sizes, shapes and positions were set to perform crack defect propagation simulation analysis. Multiple sets of balance beam crack defect propagation strain field data were obtained, including:

[0106] On the basis of the standard finite element shape function of the three-dimensional finite element simulation model of the crane balance beam, a level set function describing the crack surface and a crack tip singular function describing the stress and strain singularity at the crack tip are introduced. The level set function is a signed distance function used to describe the signed distance from any point in space to the crack surface. The crack tip singular function is constructed by the linear elastic fracture mechanics theory.

[0107] The finite difference method is used to discretize the time and space derivatives of the level set function. The extension and bifurcation of the crack surface are determined by iteratively updating the level set function value. According to the updated level set function value, the level set function value at each finite element mesh node is calculated by finite element interpolation to determine the position of the crack surface.

[0108] The relative position relationship between the crack surface and the finite element mesh is determined according to the position of the crack surface, and the crack surface enhancement degree of freedom is dynamically updated using the geometric enhancement criterion. For the unit completely cut by the crack surface, the crack surface enhancement degree of freedom is introduced at its node;

[0109] According to the size of the crack tip plastic zone, the crack tip singularity enhancement degree of freedom is dynamically updated using the plastic zone estimation criterion. The size of the crack tip plastic zone is estimated using fracture mechanics theory. For the finite element mesh contained in the crack tip plastic zone, the crack tip singularity enhancement degree of freedom is introduced at the node.

[0110] Based on the crack surface enhanced degrees of freedom and crack tip singularity enhanced degrees of freedom, the stiffness matrix and load vector of the three-dimensional finite element simulation model of the crane balance beam are locally modified to obtain the target crane balance beam finite element simulation model. The target crane balance beam finite element simulation model is used to carry out crack defect propagation simulation analysis and obtain multiple sets of strain field data of crack defect propagation of the crane balance beam.

[0111] For example, in the three-dimensional finite element simulation model of the crane balance beam, the fused extended finite element method is used to describe the crack defect propagation process. Based on the standard finite element shape function, two types of enhanced functions are introduced: the level set function describing the crack surface and the crack tip singular function describing the stress and strain singularity at the crack tip.

[0112] The level set function is a signed distance function that is used to describe the signed distance from any point in space to the crack surface. The signed distance function takes a value of zero on the crack surface, a positive value on one side of the crack surface, and a negative value on the other side. Its absolute value represents the shortest distance from the point to the crack surface. The level set function can implicitly represent the position and shape of the crack surface, avoiding the complexity of explicitly tracking the crack surface.

[0113] The crack tip singular function is constructed based on the theory of linear elastic fracture mechanics and is used to directly describe the singularity of the stress-strain field at the crack tip. By introducing a crack tip singular function that is adapted to the singularity, this singularity can be directly described in the finite element discretization, thereby improving the accuracy of the numerical solution.

[0114] In order to track the expansion and bifurcation of the crack surface, the finite difference method is used to discretize the time and space derivatives of the level set function. At each time step, the evolution of the crack surface is realized by iteratively updating the level set function value. Specifically, the time derivative and space derivative of the level set function are calculated using the finite difference method, and then the level set function value is updated according to the velocity equation of crack expansion. At the end of each iteration, the level set function value at the grid node is calculated using finite element interpolation based on the updated level set function value, and then the new position of the crack surface is determined.

[0115] According to the position of the crack surface, the relative position relationship between the crack surface and the finite element mesh can be determined. For the unit completely cut by the crack surface, the geometric enhancement criterion is used to introduce the crack surface enhanced freedom at its node. The geometric enhancement criterion means judging whether it is necessary to introduce enhanced freedom at the unit node according to the geometric relationship between the crack surface and the unit. The purpose of introducing the crack surface enhanced freedom is to capture the discontinuous displacement field on both sides of the crack surface and improve the accuracy of the numerical solution.

[0116] On the other hand, according to the size of the crack tip plastic zone, the crack tip singularity enhancement degree of freedom is dynamically updated using the plastic zone estimation criterion. The plastic zone refers to the area near the crack tip where stress concentration causes plastic deformation of the material. The size of the crack tip plastic zone can be estimated using fracture mechanics theory. For the elements contained in the plastic zone, the crack tip singularity enhancement degree of freedom is introduced at its nodes to accurately describe the stress-strain singularity at the crack tip.

[0117] After introducing the crack surface enhancement degrees of freedom and the crack tip singularity enhancement degrees of freedom, the stiffness matrix and load vector of the extended finite element method need to be locally modified. For the elements containing enhanced degrees of freedom, the terms related to the enhancement function are added to their stiffness matrix and load vector. These additional terms reflect the contribution of the enhancement function to the element stiffness and nodal force, ensuring the numerical stability and convergence of the extended finite element method.

[0118] In the fused extended finite element method, after introducing the crack surface enhanced degrees of freedom and the crack tip singularity enhanced degrees of freedom, it is necessary to locally modify the stiffness matrix and load vector of the three-dimensional finite element simulation model of the crane balance beam to obtain the target crane balance beam finite element simulation model.

[0119] First, for the elements containing crack surface enhancement degrees of freedom, add items related to the crack surface enhancement function in their stiffness matrix and load vector. Specifically, the product of the standard finite element shape function and the crack surface enhancement function is used as a new shape function, and these new shape functions are used in the calculation of the element stiffness matrix and load vector. In this way, the influence of the crack surface enhancement degrees of freedom can be reflected at the element level, capturing the discontinuous displacement field on both sides of the crack surface.

[0120] Secondly, for the unit containing the crack tip singularity enhanced degrees of freedom, add the terms related to the crack tip singular function in its stiffness matrix and load vector. Similar to the crack surface enhanced degrees of freedom, the product of the standard finite element shape function and the crack tip singular function is used as a new shape function, and these new shape functions are used in the calculation of the unit stiffness matrix and load vector. In this way, the influence of the crack tip singularity enhanced degrees of freedom can be reflected at the unit level, and the stress and strain singularity at the crack tip can be accurately described.

[0121] In this embodiment, by integrating the extended finite element method, a crack defect can be introduced into the three-dimensional finite element simulation model of the crane balance beam, and the expansion process of the crack defect under the external load can be simulated. By performing multiple sets of simulation analysis, the crack extension strain field data of the crane balance beam under different crack defect sizes, shapes and positions can be obtained, providing important numerical simulation results for the fracture failure analysis of the crane balance beam.

[0122] In an optional embodiment,

[0123] The finite element simulation model of the target crane balance beam is used to perform crack defect propagation simulation analysis, and multiple sets of strain field data of the crane balance beam crack defect propagation are obtained, including:

[0124] An initial crack defect is introduced at a preset position in the finite element simulation model of the target crane balance beam, and the signed distance value at the finite element mesh node containing the initial crack defect is calculated using the level set function. The crack surface enhancement degree of freedom is introduced into the finite element mesh node whose signed distance value is less than the preset distance threshold, so as to form a crack surface enhancement unit node set; and according to the size of the crack tip plastic zone, the crack tip singularity enhancement degree of freedom is introduced into the finite element mesh node located in the crack tip plastic zone, so as to form a crack tip singularity enhancement unit node set;

[0125] Apply loads and boundary conditions under actual working conditions to the finite element simulation model of the target crane balance beam, and select appropriate time steps and convergence criteria according to the nature of the load and the time scale of crack growth to perform crack growth analysis;

[0126] In the crack growth analysis process, for each time step, the strain field and stress field are calculated based on the node displacement field obtained in the previous time step, and the crack tip is judged whether it meets the growth condition according to the material fracture criterion. If the growth condition is met, the crack growth direction and crack growth increment are determined according to the maximum stress intensity factor principle;

[0127] The level set function value is updated according to the crack extension increment, and the signed distance value at the finite element mesh node is recalculated. For the new crack surface position, the crack surface enhancement unit node set and the crack tip singularity enhancement unit node set are updated. The iteration is repeated until the preset simulation time is reached, and finally multiple sets of strain field data of the crack defect extension of the crane balance beam are output.

[0128] For example, first, in the three-dimensional finite element simulation model, select a suitable location to introduce an initial crack defect, and the size and shape of the crack defect can be set according to the actual situation or with reference to relevant standards. Then, apply loads and boundary conditions under actual working conditions to the finite element model, including deadweight load, hoisting load, support and connection boundaries, etc. According to the nature of the load and the time scale of crack propagation, select a suitable time step and convergence criterion to ensure the stability and convergence of the numerical simulation.

[0129] In the crack growth analysis process, for each time step, the unit strain field and stress field are first calculated based on the node displacement field obtained in the previous step. Fracture mechanics parameters such as stress intensity factor or strain energy release rate are used, combined with the fracture toughness and fracture criterion of the material, to determine whether the crack tip meets the growth conditions. If the growth conditions are met, the crack growth direction and increment are determined based on the minimum energy principle or the maximum stress intensity factor principle.

[0130] According to the crack extension increment, the level set function value is updated, and the signed distance value at the unit node is recalculated. For the new crack surface position, the unit node set containing the crack surface enhanced freedom degree is updated; at the same time, according to the new crack tip plastic zone size, the unit node set containing the crack tip singularity enhanced freedom degree is updated.

[0131] Repeat the above crack growth analysis steps until the preset simulation time is reached, or the crack grows to a critical state, such as the crack runs through the entire cross section or the crack growth rate tends to be stable.

[0132] During the crack propagation analysis, the strain field and stress field data of each time step are output, including the strain distribution cloud map and stress distribution cloud map of the crack propagation area, as well as the strain stress cloud map of the entire balance beam, and finally multiple sets of strain field data of the crack defect propagation of the crane balance beam are output. Through the analysis and visualization of these data, the impact of crack defect propagation on the structural integrity and bearing capacity of the balance beam can be intuitively evaluated.

[0133] In this embodiment, through the finite element simulation model of the target crane balance beam, on the basis of the standard finite element method, the crack surface enhancement degree of freedom and the crack tip singularity enhancement degree of freedom are introduced, which can more accurately describe the stress-strain field during the crack defect expansion process. Through multiple groups of simulation analysis, the crack extension strain field data of the crane balance beam under different initial crack defect sizes, shapes and positions can be obtained. These data can be used to evaluate the impact of crack defects on the structural integrity and safety of the balance beam, and provide a basis for formulating reasonable detection and maintenance strategies. Further improve the fusion and extended finite element method so that it can more efficiently and accurately simulate the crack defect expansion process in complex structures such as crane balance beams, and provide a reliable numerical simulation tool for structural integrity assessment and life prediction.

[0134] S103. Use the crack defect deformation ultrasound-strain field fusion feature vector to train a crack defect deformation detection intelligent discrimination model, obtain the time-frequency domain feature vector and strain distribution feature vector of the crane balance beam to be detected, input them into the trained crack defect deformation detection intelligent discrimination model for intelligent discrimination screening, and output the crack defect detection results of the balance beam to be detected, including the type, size, shape and specific location of the crack defect.

[0135] In an optional embodiment,

[0136] Using the crack defect deformation ultrasound-strain field fusion feature vector to train the crack defect deformation detection intelligent discrimination model includes:

[0137] Based on the ultrasonic-strain field fusion characteristics of the crack defect deformation of the balance beam, the source domain dataset and the target domain dataset are constructed respectively, where the source domain dataset consists of labeled training samples, and the target domain dataset consists of unlabeled test samples;

[0138] Using the source domain dataset, the initial intelligent discrimination model is trained by minimizing the classification loss function on the source domain. The domain adaptation mechanism is introduced into the initial intelligent discrimination model. The conditional domain adversarial neural network method is used to achieve feature alignment between the source domain and the target domain by minimizing the domain discrimination loss function, thus obtaining a domain adaptation intelligent discrimination model.

[0139] The domain adaptation intelligent discrimination model is end-to-end trained by using source domain datasets and target domain datasets, and the domain adaptation loss function of the domain adaptation intelligent discrimination model is constructed, which includes classification loss and adversarial loss. The model parameters are updated by using an alternating optimization strategy, and the trained crack defect deformation detection intelligent discrimination model is finally obtained by minimizing the domain adaptation loss function.

[0140] Exemplarily, first, based on the deformation ultrasonic-strain field fusion characteristics of the balance beam crack defect, the source domain data set and the target domain data set are constructed respectively. Deformation ultrasound is a nondestructive testing technology based on ultrasonic guided wave detection of tiny deformations of materials, which reflects the deformation state of the material through the change of ultrasonic guided wave parameters. The strain field characterizes the spatiotemporal evolution law of strain distribution during material deformation, and is an important basis for evaluating the deformation characteristics of materials. By fusing the deformation ultrasonic signal of the balance beam crack defect with the strain field data, a fusion feature that fully reflects the deformation characteristics of the crack defect can be obtained. The source domain data set consists of labeled balance beam crack defect deformation ultrasonic-strain field fusion feature training samples, and the target domain data set consists of unlabeled balance beam crack defect deformation ultrasonic-strain field fusion feature test samples.

[0141] Using the source domain data set, the initial intelligent discrimination model is trained by minimizing the classification loss function on the source domain. The intelligent discrimination model adopts a deep neural network structure and consists of two parts: a feature extractor and a classifier. The feature extractor is responsible for extracting high-level semantic features from the input balance beam crack defect deformation ultrasonic-strain field fusion features, and the classifier is responsible for identifying and classifying crack defects based on the extracted features. By training the intelligent discrimination model on the source domain data set and minimizing the classification loss function, the intelligent discrimination model can have better crack defect discrimination performance on the source domain data. The classification loss function measures the difference between the prediction result of the intelligent discrimination model and the true label. Commonly used classification loss functions include cross entropy loss function and mean square error loss function.

[0142] The domain adaptation mechanism is introduced into the initial intelligent discriminant model, and the conditional domain adversarial neural network method is used to achieve feature alignment of the source domain and the target domain by minimizing the domain discriminant loss function, thereby obtaining a domain-adapted intelligent discriminant model. The conditional domain adversarial neural network is a domain adaptation method based on adversarial learning. It achieves feature alignment of the source domain and the target domain by introducing a domain discriminator and conditional information. The domain discriminator is responsible for distinguishing whether the sample comes from the source domain or the target domain, and the conditional information is the category label of the sample. By minimizing the domain discriminant loss function, the feature distribution of the source domain and the target domain is made as close as possible while maintaining the consistency of the category information, thereby achieving feature alignment. The domain discriminant loss function measures the discriminative ability of the domain discriminator. Commonly used domain discriminant loss functions include the cross entropy loss function and the Wasserstein distance.

[0143] The domain adaptation intelligent discrimination model is trained end-to-end using the source domain dataset and the target domain dataset. The domain adaptation loss function of the domain adaptation intelligent discrimination model is constructed, which includes two parts: classification loss and adversarial loss. The classification loss measures the classification performance of the model on the source domain dataset, and the adversarial loss measures the distribution difference between the source domain and the target domain. The model parameters are updated using an alternating optimization strategy, that is, in each training iteration, the parameters of the domain discriminator are first fixed, the parameters of the feature extractor and classifier are optimized, and the classification loss is minimized; then the parameters of the feature extractor and classifier are fixed, and the parameters of the domain discriminator are optimized to minimize the adversarial loss. Through alternating optimization, the feature extractor can not only extract features that are beneficial to crack defect discrimination, but also reduce the distribution difference between the source domain and the target domain, thereby obtaining a more robust intelligent discrimination model for crack defect deformation detection.

[0144] Optionally, the performance of the trained intelligent discrimination model for crack defect deformation detection is evaluated on the target domain dataset. Evaluation indicators such as classification accuracy, precision, recall, and F1 value are used to measure the crack defect detection performance of the intelligent discrimination model on the target domain dataset. Classification accuracy indicates the proportion of correctly classified samples to the total number of samples, precision indicates the proportion of samples correctly classified into a certain category to the number of samples predicted to be in that category, recall indicates the proportion of samples correctly classified into a certain category to the number of samples actually in that category, and F1 value is the harmonic mean of precision and recall. By comprehensively analyzing multiple evaluation indicators, the performance of the intelligent discrimination model can be comprehensively evaluated, providing a reference for practical engineering applications.

[0145] In an optional embodiment,

[0146] The calculation formula of the domain adaptation loss function for constructing the domain adaptation intelligent discrimination model is as follows:

[0147] ;

[0148] Where L represents the domain adaptation loss function, n s represents the number of training samples in the source domain dataset, l (·) represents the classification loss function, F θ (·) represents the domain adaptation intelligent discriminant model, θ represents the parameters of the domain adaptation intelligent discriminant model, represents the i-th training sample in the source domain dataset, represents the true label of the i-th training sample in the source domain dataset, λ represents the hyperparameter for balancing classification loss and adversarial loss, and k δ (·) represents the domain discriminator, δ represents the domain discriminator parameter, n t represents the number of test samples in the target domain dataset, represents the jth test sample in the target domain dataset.

[0149] In this embodiment, the generalization ability of the model between different domains (such as different working environments or different devices) is significantly improved through the domain adaptation mechanism, so that the model trained on one domain can be better applied to other domains. The use of unlabeled target domain data for training greatly reduces the dependence on a large amount of labeled data, reducing the cost and workload of data labeling. The conditional domain adversarial neural network method enables the model to automatically adapt to the differences between different domains, improving the adaptability and robustness of the model in practical applications. It not only improves the accuracy and reliability of crack defect detection, but also greatly enhances the adaptability and practicality of the model, providing a more powerful and flexible tool for the safety monitoring of crane balance beams.

[0150] Figure 2FIG. 1 is a schematic diagram of a structure of a crane balance beam crack defect detection system according to an embodiment of the present invention. Figure 2 As shown, the system comprises:

[0151] The first unit is used to arrange multiple ultrasonic guided wave sensor arrays on the surface of the balance beam of the crane to be tested, collect ultrasonic guided wave signals inside the balance beam, obtain ultrasonic guided wave signal data of the balance beam and perform wavelet time-frequency analysis to obtain a time-frequency energy distribution diagram of the ultrasonic guided wave of the balance beam, extract multi-scale and multi-directional texture features of the time-frequency energy distribution diagram of the ultrasonic guided wave of the balance beam, and construct a time-frequency domain feature vector of the ultrasonic guided wave of the balance beam;

[0152] The second unit is used to construct a three-dimensional finite element simulation model of a crane balance beam, set crack defects of different sizes, shapes and positions, perform crack defect extension simulation analysis, obtain multiple sets of balance beam crack defect extension strain field data, extract the strain distribution morphological characteristics, strain gradient characteristics and strain statistical characteristics of the multiple sets of balance beam crack defect extension strain field data, construct a balance beam crack defect strain field feature vector, fuse the balance beam ultrasonic guided wave time-frequency domain feature vector and the balance beam crack defect strain field feature vector, and construct a crack defect deformation ultrasonic-strain field fusion feature vector;

[0153] The third unit is used to train a crack defect deformation detection intelligent discrimination model using the crack defect deformation ultrasound-strain field fusion feature vector, obtain the time-frequency domain feature vector and strain distribution feature vector of the crane balance beam to be detected, input the trained crack defect deformation detection intelligent discrimination model for intelligent discrimination screening, and output the crack defect detection result of the balance beam to be detected, including the type, size, shape and specific position of the crack defect.

[0154] According to a third aspect of the embodiments of the present invention,

[0155] An electronic device is provided, comprising:

[0156] processor;

[0157] a memory for storing processor-executable instructions;

[0158] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0159] A fourth aspect of the embodiments of the present invention is:

[0160] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0161] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0162] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting crack defects in a crane balance beam, characterized in that: include: Arrange multiple ultrasonic guided wave sensor arrays on the surface of the balance beam of the crane to be tested, collect ultrasonic guided wave signals inside the balance beam, obtain ultrasonic guided wave signal data of the balance beam, perform wavelet time-frequency analysis, obtain a time-frequency energy distribution diagram of the ultrasonic guided wave of the balance beam, extract multi-scale and multi-directional texture features of the time-frequency energy distribution diagram of the ultrasonic guided wave of the balance beam, and construct a time-frequency domain feature vector of the ultrasonic guided wave of the balance beam; Construct a three-dimensional finite element simulation model of a crane balance beam, set crack defects of different sizes, shapes and positions, perform crack defect extension simulation analysis, obtain multiple sets of balance beam crack defect extension strain field data, extract the strain distribution morphological characteristics, strain gradient characteristics and strain statistical characteristics of the multiple sets of balance beam crack defect extension strain field data, construct a balance beam crack defect strain field feature vector, fuse the balance beam ultrasonic guided wave time-frequency domain feature vector and the balance beam crack defect strain field feature vector, and construct a crack defect deformation ultrasonic-strain field fusion feature vector; The crack defect deformation ultrasound-strain field fusion feature vector is used to train a crack defect deformation detection intelligent discrimination model, and the time-frequency domain feature vector and strain distribution feature vector of the crane balance beam to be detected are obtained. The trained crack defect deformation detection intelligent discrimination model is input for intelligent discrimination screening, and the crack defect detection result of the balance beam to be detected is output, including the type, size, shape and specific position of the crack defect.

2. The method according to claim 1, characterized in that Multiple ultrasonic guided wave sensor arrays are arranged on the surface of the balance beam of the crane to be tested, and the ultrasonic guided wave signals inside the balance beam are collected to obtain the ultrasonic guided wave signal data of the balance beam and perform wavelet time-frequency analysis to obtain the ultrasonic guided wave time-frequency energy distribution diagram of the balance beam, including: Add Gaussian white noise of different scales to the collected ultrasonic guided wave signal inside the balance beam and perform ensemble empirical mode decomposition to obtain a series of intrinsic mode function components, calculate the correlation coefficient between each intrinsic mode function component and the ultrasonic guided wave signal inside the balance beam, automatically identify the noise-dominated intrinsic mode function components according to a preset correlation coefficient threshold and remove them, retain the intrinsic mode function components reflecting the ultrasonic guided wave signal inside the balance beam and reconstruct them, and obtain the denoised ultrasonic guided wave signal inside the balance beam; The Hilbert transform is performed on the ultrasonic guided wave signal inside the balance beam after noise reduction to obtain an analytical signal, the envelope of the instantaneous amplitude of the analytical signal is extracted, and the local maximum point of the instantaneous amplitude is determined by using the local peak detection method, and the time interval between two adjacent local maximum points is compared with the preset peak interval threshold, and the time interval between the two local maximum points is determined as the boundary interval of different guided wave modes according to the comparison result; Adaptively dividing the ultrasonic guided wave signal inside the balance beam after noise reduction according to the boundary interval, extracting the corresponding ultrasonic guided wave modal signal segment in each boundary interval, and adaptively separating different ultrasonic guided wave modal signals; Constructing a wavelet basis function library, wherein the wavelet basis function library includes a plurality of wavelet basis function families, each wavelet basis function family is composed of wavelet basis functions of different orders and scale factors, and simultaneously extracting statistical features and frequency domain features of each separated ultrasonic guided wave modal signal segment to construct ultrasonic guided wave signal features; Calculating the correlation coefficient between each wavelet basis function in the wavelet basis function library and the ultrasonic guided wave signal characteristics at different scales to obtain a correlation coefficient matrix, constructing a wavelet basis function adaptability measurement model based on the correlation coefficient matrix, using the output of the wavelet basis function adaptability measurement model as the fitness function, and searching the wavelet basis function library for the best matching wavelet basis function using a genetic algorithm; Based on the obtained best matching wavelet basis function, an adaptive wavelet time-frequency transform is performed on the ultrasonic guided wave signal characteristics, and finally a time-frequency energy distribution diagram of the balanced beam ultrasonic guided wave is obtained.

3. The method according to claim 2, characterized in that Based on the obtained best matching wavelet basis function, the characteristics of the ultrasonic guided wave signal are subjected to adaptive wavelet time-frequency transformation, and finally the time-frequency energy distribution diagram of the balanced beam ultrasonic guided wave is obtained, including: The ultrasonic guided wave signal characteristics are subjected to wavelet transform using the best matching wavelet basis function to obtain a time-frequency representation matrix, the energy distribution discreteness of the time-frequency representation matrix at different scales is analyzed, and the scale parameter of the wavelet transform is adaptively adjusted according to the minimum energy distribution discreteness criterion to obtain the optimal scale parameter range; Within the optimal scale parameter range, the optimal translation parameter is obtained by optimizing the translation parameter of the best matching wavelet basis function and minimizing the mean square error between the spectrum of the best matching wavelet basis function and the characteristic spectrum of the ultrasonic guided wave signal; Based on the obtained best matching wavelet basis function, optimal scale parameter and optimal translation parameter, the characteristics of the ultrasonic guided wave signal are subjected to adaptive wavelet time-frequency transformation, and finally a time-frequency energy distribution diagram of the balanced beam ultrasonic guided wave is obtained; Among them, the calculation formulas of the optimal scale parameter and the optimal translation parameter are as follows: ； Among them, a * represents the optimal scale parameter, b * represents the optimal translation parameter, a represents the scale parameter, b represents the translation parameter, E(·) represents the energy distribution discreteness, μ represents the weight coefficient of the spectrum matching degree, S x (·) represents the frequency spectrum of ultrasonic guided wave signal characteristics, S Ψ (·) represents the frequency spectrum of the wavelet basis function, and f represents the frequency.

4. The method according to claim 1, characterized in that: A three-dimensional finite element simulation model of the crane balance beam was constructed, and crack defects of different sizes, shapes and positions were set to perform crack defect propagation simulation analysis. Multiple sets of balance beam crack defect propagation strain field data were obtained, including: On the basis of the standard finite element shape function of the three-dimensional finite element simulation model of the crane balance beam, a level set function describing the crack surface and a crack tip singular function describing the stress and strain singularity at the crack tip are introduced. The level set function is a signed distance function used to describe the signed distance from any point in space to the crack surface. The crack tip singular function is constructed by the linear elastic fracture mechanics theory. The finite difference method is used to discretize the time and space derivatives of the level set function. The extension and bifurcation of the crack surface are determined by iteratively updating the level set function value. According to the updated level set function value, the level set function value at each finite element mesh node is calculated by finite element interpolation to determine the position of the crack surface. The relative position relationship between the crack surface and the finite element mesh is determined according to the position of the crack surface, and the crack surface enhancement degree of freedom is dynamically updated using the geometric enhancement criterion. For the unit completely cut by the crack surface, the crack surface enhancement degree of freedom is introduced at its node; According to the size of the crack tip plastic zone, the crack tip singularity enhancement degree of freedom is dynamically updated using the plastic zone estimation criterion. The size of the crack tip plastic zone is estimated using fracture mechanics theory. For the finite element mesh contained in the crack tip plastic zone, the crack tip singularity enhancement degree of freedom is introduced at the node. Based on the crack surface enhanced degrees of freedom and crack tip singularity enhanced degrees of freedom, the stiffness matrix and load vector of the three-dimensional finite element simulation model of the crane balance beam are locally modified to obtain the target crane balance beam finite element simulation model. The target crane balance beam finite element simulation model is used to carry out crack defect propagation simulation analysis and obtain multiple sets of strain field data of crack defect propagation of the crane balance beam.

5. The method according to claim 4, characterized in that The finite element simulation model of the target crane balance beam is used to perform crack defect propagation simulation analysis, and multiple sets of strain field data of the crane balance beam crack defect propagation are obtained, including: An initial crack defect is introduced at a preset position in the finite element simulation model of the target crane balance beam, and the signed distance value at the finite element mesh node containing the initial crack defect is calculated using the level set function. The crack surface enhancement degree of freedom is introduced into the finite element mesh node whose signed distance value is less than the preset distance threshold, so as to form a crack surface enhancement unit node set; and according to the size of the crack tip plastic zone, the crack tip singularity enhancement degree of freedom is introduced into the finite element mesh node located in the crack tip plastic zone, so as to form a crack tip singularity enhancement unit node set; Apply loads and boundary conditions under actual working conditions to the finite element simulation model of the target crane balance beam, and select appropriate time steps and convergence criteria according to the nature of the load and the time scale of crack growth to perform crack growth analysis; In the crack growth analysis process, for each time step, the strain field and stress field are calculated based on the node displacement field obtained in the previous time step, and the crack tip is judged whether it meets the growth condition according to the material fracture criterion. If the growth condition is met, the crack growth direction and crack growth increment are determined according to the maximum stress intensity factor principle; The level set function value is updated according to the crack extension increment, and the signed distance value at the finite element mesh node is recalculated. For the new crack surface position, the crack surface enhancement unit node set and the crack tip singularity enhancement unit node set are updated. The iteration is repeated until the preset simulation time is reached, and finally multiple sets of strain field data of the crack defect extension of the crane balance beam are output.

6. The method according to claim 1, characterized in that Using the crack defect deformation ultrasound-strain field fusion feature vector to train the crack defect deformation detection intelligent discrimination model includes: Based on the ultrasonic-strain field fusion characteristics of the crack defect deformation of the balance beam, the source domain dataset and the target domain dataset are constructed respectively, where the source domain dataset consists of labeled training samples, and the target domain dataset consists of unlabeled test samples; Using the source domain dataset, the initial intelligent discrimination model is trained by minimizing the classification loss function on the source domain. The domain adaptation mechanism is introduced into the initial intelligent discrimination model. The conditional domain adversarial neural network method is used to achieve feature alignment between the source domain and the target domain by minimizing the domain discrimination loss function, thus obtaining a domain adaptation intelligent discrimination model. The domain adaptation intelligent discrimination model is end-to-end trained by using source domain datasets and target domain datasets, and the domain adaptation loss function of the domain adaptation intelligent discrimination model is constructed, which includes classification loss and adversarial loss. The model parameters are updated by using an alternating optimization strategy, and the trained crack defect deformation detection intelligent discrimination model is finally obtained by minimizing the domain adaptation loss function.

7. The method according to claim 6, characterized in that The calculation formula of the domain adaptation loss function for constructing the domain adaptation intelligent discrimination model is as follows: ； Where L represents the domain adaptation loss function, n s represents the number of training samples in the source domain dataset, l (·) represents the classification loss function, F θ (·) represents the domain adaptation intelligent discriminant model, θ represents the parameters of the domain adaptation intelligent discriminant model, represents the i-th training sample in the source domain dataset, represents the true label of the i-th training sample in the source domain dataset, λ represents the hyperparameter for balancing classification loss and adversarial loss, and k δ (·) represents the domain discriminator, δ represents the domain discriminator parameter, n t represents the number of test samples in the target domain dataset, represents the jth test sample in the target domain dataset.

8. A crane balance beam crack defect detection system, used to implement the method described in any one of claims 1 to 7, characterized in that: include: The first unit is used to arrange multiple ultrasonic guided wave sensor arrays on the surface of the balance beam of the crane to be tested, collect ultrasonic guided wave signals inside the balance beam, obtain ultrasonic guided wave signal data of the balance beam and perform wavelet time-frequency analysis to obtain a time-frequency energy distribution diagram of the ultrasonic guided wave of the balance beam, extract multi-scale and multi-directional texture features of the time-frequency energy distribution diagram of the ultrasonic guided wave of the balance beam, and construct a time-frequency domain feature vector of the ultrasonic guided wave of the balance beam; The second unit is used to construct a three-dimensional finite element simulation model of a crane balance beam, set crack defects of different sizes, shapes and positions, perform crack defect extension simulation analysis, obtain multiple sets of balance beam crack defect extension strain field data, extract the strain distribution morphological characteristics, strain gradient characteristics and strain statistical characteristics of the multiple sets of balance beam crack defect extension strain field data, construct a balance beam crack defect strain field feature vector, fuse the balance beam ultrasonic guided wave time-frequency domain feature vector and the balance beam crack defect strain field feature vector, and construct a crack defect deformation ultrasonic-strain field fusion feature vector; The third unit is used to train a crack defect deformation detection intelligent discrimination model using the crack defect deformation ultrasound-strain field fusion feature vector, obtain the time-frequency domain feature vector and strain distribution feature vector of the crane balance beam to be detected, input the trained crack defect deformation detection intelligent discrimination model for intelligent discrimination screening, and output the crack defect detection result of the balance beam to be detected, including the type, size, shape and specific position of the crack defect.

9. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method described in any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images