Defect detection method and apparatus based on nonlinear system identification

Abstract

The present invention relates to the technical field of system identification. Provided are a defect detection method and apparatus based on nonlinear system identification. The method comprises: using a modal impact hammer to perform hammer impact excitation on a test piece according to a test schedule, so as to generate excitation signals and response signals; using a laser vibrometer to collect the excitation signals and the response signals; performing parameter calculation on the excitation signals and the response signals to obtain model parameters, constructing an initial Hammerstein model on the basis of the model parameters, and optimizing the initial Hammerstein model to obtain a test piece Hammerstein model; and using a cross validation method to adjust the test piece Hammerstein model, comparing the adjusted test piece Hammerstein model with a preconstructed template test piece model for defect determination, and on the basis of a determination result, determining whether the test piece has defects. As defects of a test piece induce nonlinear effects, signals generated by the nonlinear effects are used to establish a nonlinear model of the test piece, so as to determine on the basis of the nonlinear model whether the test piece has defects.

Description

Defect detection method and device for nonlinear system identification TECHNICAL FIELD

[0001] The present application mainly relates to the technical field of system identification, and particularly relates to a defect detection method and device for nonlinear system identification. BACKGROUND

[0002] In the production and use of metal materials, defects in metal materials are a common problem. If metal parts are made of metal materials and have defects, the structure of the metal parts will be damaged, thereby affecting the normal operation and safety of the equipment. Therefore, metal material defect detection is a very important detection work in the metal industry field. In the detection process, it is also necessary to ensure that the detection object is not damaged to avoid increasing defects, so that the detection object is non-destructively detected.

[0003] In the field of system identification, system identification is carried out through linear models and nonlinear models, and the parameter identification technology for linear models is relatively mature. However, if the state and output variables of a system cannot be described by a linear relationship under the influence of external conditions, the system belongs to a nonlinear system. Since metal materials have defects, nonlinear effects are generated, but currently the combination of system identification technology and non-destructive detection technology and the research into actual production and detection are still in the initial stage. Therefore, how to combine system identification technology with non-destructive detection technology has become a difficult problem to be solved. SUMMARY

[0004] The technical problem to be solved by the present application is to combine non-destructive detection technology with nonlinear system identification technology to detect defects in metal materials, and to provide a defect detection method and device for nonlinear system identification.

[0005] The technical solution of the present application to solve the above technical problem is as follows: a defect detection method for nonlinear system identification, comprising the following steps:

[0006] A test plan table is set according to the size of a test piece, a modal force hammer is used to hammer the test piece according to the test plan table to generate an excitation signal, the test piece is vibrated based on the excitation signal to generate a response signal;

[0007] A laser vibration meter is used to collect the excitation signal and the response signal respectively;

[0008] The excitation signal and the response signal are subjected to nonlinear parameter calculation to obtain initial nonlinear parameters, the excitation signal and the response signal are subjected to linear parameter calculation to obtain initial linear parameters, an initial Hammerstein model is constructed according to the initial nonlinear parameters and the initial linear parameters, and the initial Hammerstein model is optimized to obtain a test piece Hammerstein model.

[0009] The test piece Hammerstein model is adjusted through a cross-validation method, the adjusted test piece Hammerstein model is subjected to defect judgment together with a template test piece model constructed in advance, and whether the test piece has defects is determined according to a judgment result.

[0010] Another technical solution of the present application for solving the above technical problem is as follows:

[0011] A defect detection device for nonlinear system identification, comprising a signal generation unit, a signal acquisition unit, a model construction unit and a defect detection unit.

[0012] The signal generation unit is configured to set a test schedule according to a test piece size, to make a hammering excitation on the test piece according to the test schedule through a modal force hammer to generate an excitation signal, and to make the test piece vibrate based on the excitation signal to generate a response signal.

[0013] The signal acquisition unit is configured to acquire the excitation signal and the response signal through a laser vibration meter.

[0014] The model construction unit is configured to perform nonlinear parameter calculation on the excitation signal and the response signal to obtain initial nonlinear parameters, to perform linear parameter calculation on the excitation signal and the response signal to obtain initial linear parameters, to construct an initial Hammerstein model according to the initial nonlinear parameters and the initial linear parameters, and to optimize the initial Hammerstein model to obtain a test piece Hammerstein model.

[0015] The defect detection unit is configured to adjust the test piece Hammerstein model through a cross-validation method, to perform defect judgment on the adjusted test piece Hammerstein model together with a template test piece model constructed in advance, and to determine whether the test piece has defects according to a judgment result.

[0016] The present application has the following beneficial effects: the present application optimizes the structure damage detection technology for test pieces, and makes it possible to identify defects in the form of establishing a nonlinear system model.

[0017] The system identification technology is used for nondestructive testing of the test piece. Nonlinear effects are caused by micro-defects on the test piece. A Hammerstein model is selected as the nonlinear model. The introduced to-be-identified parameters are less and the flexibility is strong. The model corresponding to the test piece is compared with a template model. The model characteristics are summarized. Qualitative analysis is performed. The purpose of defect detection is achieved. BRIEF DESCRIPTION OF DRAWINGS

[0018] Fig. 1 is a flowchart of a defect detection method of nonlinear system identification provided by an embodiment of the present application;

[0019] Fig. 2 is a structural schematic diagram of a defect detection device of nonlinear system identification provided by an embodiment of the present application;

[0020] Fig. 3 is a unit block diagram of a defect detection device of nonlinear system identification provided by an embodiment of the present application.

[0021] In the drawings, the component names represented by each mark are as follows:

[0022] 1, test piece; 2, modal force hammer; 3, laser vibration meter; 4, signal processing platform. DETAILED DESCRIPTION

[0023] The principles and characteristics of the present application are described below in combination with the drawings. The examples are only used to explain the present application and are not used to limit the scope of the present application.

[0024] Metal material defect detection technology is applied to the fields of aerospace, petrochemical industry and automobile manufacturing. Therefore, it is necessary to detect defects of metal materials. In the detection process, it is also necessary to ensure that the detection object is not damaged to avoid increasing defects. Therefore, nondestructive testing is performed on the detection object. However, the combination of system identification technology and nondestructive testing technology and the research on the combination of system identification technology and nondestructive testing technology in actual production and detection are still in the initial stage. How to combine the system identification technology and the nondestructive testing technology has become a difficult problem to be solved.

[0025] As shown in Fig. 1, a defect detection method of nonlinear system identification provided by an embodiment of the present application includes the following steps:

[0026] A test plan table is set according to the size of the test piece. The test piece is hammered and excited by a modal force hammer according to the test plan table. An excitation signal is generated. The test piece is vibrated based on the excitation signal. A response signal is generated.

[0027] The excitation signal and the response signal are collected by a laser vibration meter.

[0028] performing nonlinear parameter calculation on the excitation signal and the response signal to obtain initial nonlinear parameters, performing linear parameter calculation on the excitation signal and the response signal to obtain initial linear parameters, constructing an initial Hammerstein model according to the initial nonlinear parameters and the initial linear parameters, and optimizing the initial Hammerstein model to obtain a specimen Hammerstein model;

[0029] adjusting the specimen Hammerstein model through a cross-validation method, performing defect judgment on the adjusted specimen Hammerstein model and a pre-constructed template specimen model, and determining whether the specimen has defects according to a judgment result.

[0030] Specifically, as shown in FIG. 2, a common aluminum alloy sheet with a length of 250 mm, a width of 50 mm and a thickness of 2 mm is selected as a specimen, a micro crack is made on the specimen 1, and a clamp is used to fix the specimen. A modal force hammer 2 is used to apply an exciting force to the specimen according to the test schedule, so that the specimen 1 vibrates. The force signal generated by the force is the excitation signal of the nonlinear system in which the specimen 1 is located, and the vibration displacement function (also referred to as a response function) of the specimen 1 under the corresponding excitation signal is used to obtain a response signal. After all test groups in the test schedule are completed, the response signal and the excitation signal can be derived, and the vibration of the specimen 1 at each time period is processed in real time through a VSM-TEST signal processing platform (i.e., a signal processing platform 4).

[0031] Each time period can be understood as follows: since the hammering positions are different according to the test schedule, the time periods of hammering are different, and thus the specimen 1 vibrates at multiple time periods (i.e., the test time corresponding to multiple test groups).

[0032] The laser vibration meter 3 includes a laser controller and a control module, wherein the control module includes an acquisition card. The related parameters of the laser vibration meter 3 are set through the control module of the laser vibration meter 3, and the related parameters include an acquisition channel, a sampling frequency, a trigger setting and a preprocessing function, etc.

[0033] The laser vibration meter 3 is used to acquire the excitation signal and the response signal, and a laser vibration meter of a Julight brand is selected.

[0034] Before the step of setting the test schedule according to the size of the specimen, the following steps are further included.

[0035] The size, material and metal plate defect range of the specimen 1 are determined as setting parameters, the laser focal length is set through the Vibro Remote Console laser controller of the laser vibration meter 3 according to the setting parameters, and the laser is turned on.

[0036] Specifically, according to the basic data of the force hammer (i.e. modal force hammer), the basic parameters (i.e. relevant parameters) of the signal channel of the control module of the laser vibration meter 3 are set. Among them, channel one is set as an input channel, and its sensitivity is 2.42 mV / N; channel two is set as an output channel; and according to the size, material and metal plate defect range of the test piece, the sampling frame rate is set to 5.12 KHz, the spectral line is selected to be 1600, and the frequency width is 2 KHz; the linear average mode is used for excitation, the trigger delay is set to -20 ms, the trigger threshold is set to 0.4%, and the trigger hysteresis is set to 0.2%; the input signal (i.e. excitation signal) is added with a rectangular window function, and the output signal (i.e. response signal) is added with an exponential window function, the main window offset is 5 ms, and the decay constant is 100 ms; a high-pass filter is applied to the pre-processing stage, the 3dB frequency 1 is 5 Hz, and the 3dB frequency 2 is 1.2 KHz.

[0037] Further including, the direction of the laser emitted by the laser controller of the laser vibration meter 3 is set as the positive direction of the Z-axis of the world coordinate system, since the excitation signal is the excitation force applied to the test piece 1 for causing the vibration of the test piece 1, the direction of the excitation signal is set as the -Z-axis direction (i.e. opposite to the direction of the laser emission) in the setting of the test schedule; since the response signal is the signal representation of the vibration displacement function of the test piece 1 after being excited, the direction of the response signal is set as the +Z-axis direction (i.e. the same as the direction of the laser emission) in the setting of the test schedule. By turning on the laser (i.e. the laser in the positive direction of the Z-axis) through the laser controller, the modal force hammer 2 performs hammering excitation on the test piece 1 according to the test schedule, generates the excitation signal and sends it to the control module; after the excitation signal is applied to the test piece 1, the test piece 1 vibrates based on the excitation signal, and the vibration causes a slight change in the laser, the laser controller receives the slight change (i.e. the generated response signal) caused by the vibration and sends it to the control module; the excitation signal and the response signal are collected by the control module of the laser vibration meter 3, and the collected excitation signal and response signal are transmitted to the signal processing platform 4; the excitation signal and the response signal are processed by the signal processing platform 4, an initial Hammerstein model, a test piece Hammerstein model and a judgment on whether the test piece 1 has defects are constructed.

[0038] It should be understood that system identification is to determine a model (e.g. linear model or nonlinear model) equivalent to the observed system from the input data and output data of the observed system (i.e. system identification is a modeling technology using the input data and output data of the system), wherein the observed system is a test system that generates corresponding signals (i.e. excitation signal and response signal) by hammering excitation of the test piece according to the test schedule.

[0039] Non-destructive testing is carried out by a modal force hammer, and the biggest advantage of non-destructive testing is that accurate data can be obtained without damaging the measured object, so as to evaluate the performance, damage degree and other information of the measured object.

[0040] In the embodiment of the application, the model corresponding to the nonlinear system is optimized and improved in the nonlinear system identification technology, that is, the model parameter optimization algorithm is improved, and the superiority and performance of the model are improved. Referring to the hammering method in modal analysis, the metal material is excited by hammering through a modal force hammer to achieve the purpose of non-destructive testing. The laser vibration meter is used to collect and process the excitation signal and the response signal, and then the input data and the output data (i.e. the excitation signal and the response signal) of the nonlinear system of the test piece can be obtained. Since the defects of the test piece will produce nonlinear effects, the excitation signal and the response signal collected after the test piece is hammered are processed to build a nonlinear model of the test piece, so as to determine whether the test piece has defects according to the nonlinear model.

[0041] Preferably, the test plan table is set according to the size of the test piece, specifically:

[0042] The number of detection contacts is selected according to the size of the test piece, and the arrangement mode of the detection contacts is set, and the test plan table is set according to the arrangement mode.

[0043] It should be understood that the number and arrangement mode of the detection contacts of the laser vibration meter on the test piece to be detected are selected according to the size of the test piece, and then the test plan table is set, and the response signal and the excitation signal of each period are collected by the control module of the laser vibration meter according to the test order in the test plan table.

[0044] The test plan table can be used to apply multiple (i.e. different frequencies) excitation signals to the same test piece to obtain multiple corresponding response signals. The accuracy of the model can be greatly improved, and the randomness of the experiment can be fully avoided.

[0045] In the embodiment of the application, the hammering detection contacts are distributed according to the size of the metal test piece, which can detect the defects of the metal test piece in all directions, avoiding partial detection, which leads to incomplete detection of the metal test piece defects and the risk of existing micro-cracks.

[0046] Preferably, after the laser vibration meter collects the excitation signal and the response signal, respectively, the method further comprises:

[0047] The excitation signal is added with a rectangular window function, and the response signal is added with an exponential window function.

[0048] The excitation signal added with the rectangular window function and the response signal added with the exponential window function are respectively subjected to noise reduction processing.

[0049] It should be understood that the window function is a mathematical function for signal processing and spectrum analysis, and the corresponding window function is added to the excitation signal and the response signal, respectively, a window is applied to the original signal, so that the both ends of the original signal gradually attenuate, thereby reducing the influence degree of spectrum leakage (i.e., reducing spectrum leakage and improving spectrum resolution), so as to better analyze and process. Among them, the rectangular window function has the highest advantage of frequency identification accuracy; the exponential window function can smooth the sample points after weighting the signal, so that the signal becomes more continuous.

[0050] In the embodiment of the application, adding the window function to the signal helps to reduce spectrum leakage and suppress signal noise, and also helps to improve frequency resolution and reduce or eliminate noise through noise reduction processing to restore the clarity of the original signal and reduce errors during signal processing.

[0051] Preferably, the non-linear parameter calculation of the excitation signal and the response signal obtains an initial non-linear parameter, specifically:

[0052] The non-linear parameter calculation of the excitation signal and the response signal obtains an initial non-linear parameter through a polynomial non-linear function, and the polynomial non-linear function is:

[0053] p(x)=a0+a1x+a2x2+...+a n-1 x n-1 +a n x n ,

[0054] Wherein, p(x) represents an initial non-linear parameter (i.e., a parameter data corresponding to the response signal generated by the vibration of the excitation signal on the test piece), x represents a static excitation parameter (i.e., a parameter data corresponding to the excitation signal), n represents the highest degree of the polynomial, a represents the polynomial coefficient, and a>0, a≠1.

[0055] It should be understood that the polynomial non-linear function is used to describe the non-linear relationship between the input variable (i.e., the excitation signal) and the output variable (i.e., the response signal), which approximates complex non-linear phenomena through a series of power terms of input variables, and the expression form is the weighted sum of input variables.

[0056] In the embodiment of the present application, the polynomial nonlinear function has relatively low calculation complexity, and even high-order polynomial can be calculated quickly, which makes it possible to quickly calculate the parameters in a nonlinear system. Moreover, the polynomial nonlinear function can be combined with other types of models (such as a linear dynamic model) to form a complex hybrid model (i.e. a Hammerstein model) to describe the defect condition of a test piece.

[0057] Preferably, the linear parameter calculation is performed on the excitation signal and the response signal to obtain initial linear parameters, and specifically:

[0058] S1, performing linear parameter calculation on the excitation signal by a polynomial input function to obtain input linear parameters, wherein the polynomial input function is:

[0059]

[0060] wherein A(q -1 ) represents the input linear parameters, q -1 represents a unit delay operator, represents input polynomial coefficients, and n a represents the order of the input polynomial.

[0061] S2, performing linear parameter calculation on the response signal by a polynomial output function to obtain output linear parameters, wherein the polynomial output function is:

[0062]

[0063] wherein B(q -1 ) represents the output linear parameters, q -1 represents a unit delay operator, represents output polynomial coefficients, and n b represents the order of the output polynomial.

[0064] S3, calculating the input linear parameters and the output linear parameters by a discrete transfer function to obtain initial linear parameters, wherein the discrete transfer function is:

[0065]

[0066] wherein G(q -1 ) represents the initial linear parameters.

[0067] It should be understood that the unit shift operator represents a delay of a signal in time, and its function is to delay the time index of a signal sequence or function by one unit; in discrete-time signal processing, the unit shift operator can capture the change of the signal over time. The polynomial input function is the change function of the input signal (i.e. the excitation signal) after the unit shift operator calculation. The polynomial output function is the change function of the output signal (i.e. the response signal) after the unit shift operator calculation. The discrete transfer function is used to describe the linear relationship between the input signal and the output signal in discrete time, which is usually represented by a difference equation, and describes the corresponding manner of the nonlinear system to the input signal.

[0068] In the embodiment of the present application, the linear relationship between the input signal and the output signal in discrete time is calculated by the difference equation, the discrete transfer function is used for linearization analysis of the nonlinear system, and the dynamic response change of the test piece in the nonlinear system can be more accurately analyzed, and then the defect condition of the test piece is described.

[0069] Preferably, the initial Hammerstein model is optimized to obtain a test piece Hammerstein model, specifically:

[0070] The initial nonlinear parameters are calculated by the APSO algorithm to obtain nonlinear parameters;

[0071] The initial linear parameters are fitted by the least square method to obtain linear parameters;

[0072] The test piece Hammerstein model is constructed according to the nonlinear parameters and the linear parameters.

[0073] It should be understood that the essence of the parameter identification of the Hammerstein nonlinear model based on the APSO algorithm is to convert the parameter identification problem into a parameter space optimization problem, search the entire static nonlinear parameter domain in the model, and finally obtain the optimal parameters.

[0074] In the embodiment of the present application, the APSO algorithm is used to perform optimization calculation on the parameters of the static nonlinear part of the model, the individual local information and the global information of the group are used for parameter search, the convergence speed of the algorithm is improved, the requirements for the computer memory and the CPU are not high, the optimal static parameters are extracted while the identification speed is faster, that is, the difference between the identification model (i.e., the initial Hammerstein model) and the actual model (i.e., the test piece Hammerstein model) is minimized, and the model is more accurate. The least square method is used to perform fitting calculation on the parameters of the dynamic linear part of the model, the fitting process is simple, the requirement for the calculation complexity is low, the data fitting process can be completed in a short time, which is convenient and fast, and the error can be automatically corrected according to the actual situation, so that a more accurate fitting result is obtained. By adjusting the parameters calculated by the discrete transfer function, the dynamic response and the steady-state error of the system can be optimized, so that the difference between the identification model and the actual model is minimized while the identification speed is faster, and the model is more accurate.

[0075] Preferably, after the acquisition steps of the excitation signal and the response signal by the laser vibration meter, the method further comprises:

[0076] The test piece is hammered by the modal force hammer according to the test schedule to generate an excitation signal, the test piece is vibrated based on the excitation signal to generate a response signal;

[0077] The excitation signal and the response signal are acquired by the laser vibration meter;

[0078] The excitation signal and the response signal are used as a group of data, the above operation is repeated multiple times, multiple groups of data are acquired, and a data set is constructed according to the multiple groups of data.

[0079] In the embodiment of the present application, the data set is constructed, so as to adjust the test piece Hammerstein model by K-fold cross validation.

[0080] Preferably, the test piece Hammerstein model is adjusted by the cross validation method, specifically:

[0081] The pre-constructed data set is divided into multiple training sets according to the same proportion by the K-fold cross validation method, one of the training sets is selected as a validation set, the test piece Hammerstein model is trained by the remaining training sets, the test piece Hammerstein model after training is verified by the validation set, a performance evaluation index is obtained, and the above operation is repeated, each training set is used as a validation set, and the performance evaluation indexes corresponding to all validation sets are obtained.

[0082] An average value of the performance evaluation indicators corresponding to all validation sets is calculated to obtain a performance evaluation average value, and the test piece Hammerstein model is adjusted according to the performance evaluation average value.

[0083] Specifically, the pre-constructed data set is divided into K similar-sized subsets, called folds, by using the K-fold cross-validation method. Then, K times of training and validation are performed, each time selecting one fold as the validation set and the remaining K-1 folds as the training set. This process is repeated K times to ensure that each fold is used as a validation set once. Finally, the performance indicators of the K validations are averaged to obtain the performance evaluation result of the model. The parameters of the test piece model are adjusted according to the performance evaluation result to improve the fitting performance of the model.

[0084] It should be understood that the cross-validation method is a validation technique for evaluating the performance of data mining models.

[0085] In the embodiments of the present application, the effectiveness of the final model is verified by the cross-validation method, and the generalization ability of the model is evaluated by simulating the performance of the model on unknown data. The data set can be more fully utilized, the performance fluctuations of the model caused by different data divisions can be reduced, the performance of the model under different parameters can be compared to select the optimal parameters, and whether the model has overfitting or underfitting problems can be detected to evaluate the model performance and select the best model parameter characteristics.

[0086] Preferably, the adjusted test piece Hammerstein model is compared with the pre-constructed template test piece model for defect judgment, and whether the test piece has a defect is determined according to the judgment result, specifically:

[0087] The adjusted test piece Hammerstein model is compared with the pre-constructed template test piece model for feature comparison to obtain comparison features, and whether the test piece has a defect is determined according to the comparison features and a preset defect condition.

[0088] In the embodiments of the present application, the intact test piece model and the defective test piece model are compared to induce features, and the purpose of defect recognition is achieved by observing the difference between the expressions of the defective test piece model and the intact test piece model.

[0089] As shown in FIG. 3, the defect detection device for nonlinear system identification provided by the embodiments of the present application comprises a signal generation unit, a signal acquisition unit, a model construction unit and a defect detection unit.

[0090] The signal generation unit is configured to set a test schedule according to the size of the test piece, make the test piece vibrate by modal force hammer according to the test schedule to generate an excitation signal, and generate a response signal based on the excitation signal.

[0091] The signal acquisition unit is configured to acquire the excitation signal and the response signal by the laser vibration meter respectively.

[0092] The model construction unit is configured to perform nonlinear parameter calculation on the excitation signal and the response signal to obtain initial nonlinear parameters, perform linear parameter calculation on the excitation signal and the response signal to obtain initial linear parameters, construct an initial Hammerstein model according to the initial nonlinear parameters and the initial linear parameters, and optimize the initial Hammerstein model to obtain a test piece Hammerstein model.

[0093] The defect detection unit is configured to adjust the test piece Hammerstein model by a cross-validation method, perform defect judgment on the adjusted test piece Hammerstein model and a template test piece model constructed in advance, and determine whether the test piece has defects according to a judgment result.

[0094] The above-described defect detection device for nonlinear system identification can refer to the implementation content and beneficial effects of the above-described defect detection method for nonlinear system identification, which will not be described here again.

[0095] It should be noted that, in this document, the relationship terms such as first and second are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is any such actual relationship or order between these entities or operations. Moreover, the terms "include", "contain" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or device.

[0096] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described device and unit can refer to the corresponding process in the foregoing method embodiments, which will not be described here again.

[0097] In several embodiments provided in the present application, it should be understood that the disclosed device and method can be implemented by other ways. For example, the above-described device embodiments are only schematic, for example, the division of units is only a logical function division, and actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed.

[0098] The units described as separate components may or may not be physically separate, and the components displayed as units may or may not be physical units, i.e. may be located in one place, or may also be distributed to multiple network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiment of the present application.

[0099] The above description is only the preferred embodiment of the present application, and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A defect detection method for nonlinear system identification, characterized by, The method comprises the following steps: a test schedule is set according to the size of the test piece, the test piece is excited by hammering according to the test schedule by a modal force hammer, an excitation signal is generated, the test piece is vibrated based on the excitation signal, and a response signal is generated; the excitation signal and the response signal are collected by a laser vibration meter respectively; initial nonlinear parameters are calculated from the excitation signal and the response signal, initial linear parameters are calculated from the excitation signal and the response signal, an initial Hammerstein model is constructed according to the initial nonlinear parameters and the initial linear parameters, and the initial Hammerstein model is optimized to obtain a test piece Hammerstein model; the test piece Hammerstein model is adjusted by a cross-validation method, the adjusted test piece Hammerstein model is compared with a pre-constructed template test piece model to judge defects, and whether the test piece has defects is determined according to the judgment result; the adjusted test piece Hammerstein model is compared with the pre-constructed template test piece model to judge defects, and whether the test piece has defects is determined according to the judgment result, specifically as follows: the adjusted test piece Hammerstein model is compared with the pre-constructed template test piece model to judge defects, and whether the test piece has defects is determined according to the judgment result, specifically as follows:

2. The defect detection method of claim 1, wherein the number of detection contacts is selected according to the size of the test piece, the arrangement mode of the detection contacts is set, and the test schedule is set according to the arrangement mode. After the step of collecting the excitation signal and the response signal by the laser vibration meter, the following steps are further included:

3. The defect detection method of claim 1, wherein a rectangular window function is added to the excitation signal, and an exponential window function is added to the response signal; the excitation signal to which the rectangular window function is added and the response signal to which the exponential window function is added are processed for noise reduction respectively. The initial nonlinear parameters are calculated from the excitation signal and the response signal by a polynomial nonlinear function, and the polynomial nonlinear function is as follows:

4. The defect detection method of claim 1, wherein The initial linear parameters are calculated from the excitation signal and the response signal by a linear function, and the linear function is as follows: The initial Hammerstein model is optimized to obtain the test piece Hammerstein model, specifically as follows: p(x) = a0+ a1x + a2x2+... + anxn n-1 x n-1 +a n x n , the initial nonlinear parameters are calculated by an APSO algorithm to obtain nonlinear parameters; 5. The defect detection method of claim 1, wherein the initial linear parameters are fitted by a least square method to obtain linear parameters; S1, linearly parameterizing the excitation signal by a polynomial input function to obtain an input linear parameter, the polynomial input function being: where A(q -1 ) denotes the input linear parameter, q -1 denotes the unit delay operator, denote the input polynomial coefficients, n a denotes the degree of the input polynomial; S2, linear parameter calculation is performed on the response signal by a polynomial output function to obtain an output linear parameter, the polynomial output function is: where B(q -1 ) denotes the output linear parameter, q -1 denotes the unit delay operator, denotes the output polynomial coefficients, n b denotes the order of the output polynomial; S3, calculating the initial linear parameters from the input linear parameters and the output linear parameters by a discrete transfer function, the discrete transfer function being: where G(q -1 ) denotes the initial linear parameters.

6. The defect detection method of claim 1, wherein the test piece Hammerstein model is constructed according to the nonlinear parameters and the linear parameters. ​ ​ ​ 7. The defect detection method of claim 1, wherein The Hammerstein model of the test piece is adjusted through a cross-validation method, and the adjustment specifically includes: A pre-constructed data set is divided into multiple training sets in the same proportion through a K-fold cross-validation method, one of the training sets is selected as a validation set, the remaining training sets are used to train the Hammerstein model of the test piece, the trained Hammerstein model of the test piece is verified through the validation set, and a performance evaluation index is obtained, and the above process is repeated to obtain the performance evaluation index corresponding to each validation set; The performance evaluation indexes corresponding to all validation sets are averaged to obtain a performance evaluation average, and the Hammerstein model of the test piece is adjusted according to the performance evaluation average.

8. A defect detection apparatus of nonlinear system identification characterized by, It includes: A signal generation unit, a signal acquisition unit, a model construction unit and a defect detection unit; The signal generation unit is configured to set a test schedule according to the size of a test piece, and a modal force hammer is used to hammer the test piece according to the test schedule to generate an excitation signal, and the test piece is vibrated based on the excitation signal to generate a response signal; The signal acquisition unit is configured to acquire the excitation signal and the response signal through a laser vibration meter respectively; The model construction unit is configured to calculate initial nonlinear parameters of the excitation signal and the response signal to obtain initial nonlinear parameters, calculate initial linear parameters of the excitation signal and the response signal to obtain initial linear parameters, construct an initial Hammerstein model according to the initial nonlinear parameters and the initial linear parameters, and optimize the initial Hammerstein model to obtain a test piece Hammerstein model; The defect detection unit is configured to adjust the test piece Hammerstein model through a cross-validation method, compare the adjusted test piece Hammerstein model with a pre-constructed template test piece model to determine whether the test piece has defects according to a judgment result. In the defect detection unit, the adjusted test piece Hammerstein model is compared with the pre-constructed template test piece model to determine whether the test piece has defects according to a judgment result, and the adjustment specifically includes: The adjusted test piece Hammerstein model is compared with the pre-constructed template test piece model to obtain comparison features, and it is determined whether the test piece has defects according to the comparison features and a pre-set defect condition.

9. The defect detection apparatus according to claim 8, characterized by In the signal generation unit, the test schedule is set according to the size of the test piece, and the adjustment specifically includes: The number of detection contacts is selected according to the size of the test piece, the arrangement mode of the detection contacts is set, and the test schedule is set according to the arrangement mode.

Images