Component strain measurement in-situ calibration system and method based on target surface grating

Abstract

The invention provides a component strain measurement in-situ calibration system and method based on a target surface grating, the system comprises an FBG sensor and a target surface grating which are arranged at the same position of a component, the FBG sensor is connected to an FBG demodulator, the FBG demodulator demodulates the center wavelength of the FBG sensor, and the center wavelength of the FBG sensor is obtained; obtaining the central wavelength offset of the FBG sensor of the component under different strain conditions; a laser emitting device is arranged on one side of the target surface grating, a light spot collecting device is arranged on the other side of the target surface grating, the laser emitting device emits laser into the target surface grating at a preset angle, and the light spot collecting device collects diffraction light spots of the target surface grating on a focal plane. Therefore, the offset of the diffraction spot of the target surface grating on the focal plane of the component under different strain conditions can be obtained, the reference strain of the component can be calculated, and the in-situ calibration of the FBG sensor can be realized. The strain measurement precision and reliability are remarkably improved, and the method is suitable for long-term monitoring of large components and in-situ calibration of large-scale sensors.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of optical fiber sensing, and in particular to a component strain measurement in-situ calibration system and method based on a target facet grating. BACKGROUND

[0002] In the strain measurement of large components, fiber Bragg grating (FBG) sensors are widely used due to their advantages of anti-electromagnetic interference, corrosion resistance, small size, etc. However, in actual application, the measurement accuracy of FBG sensors is affected by various factors, including the thickness of the adhesive layer itself, the installation angle, environmental changes, and the strain transmission efficiency between the sensor and the base, etc. These factors cause the drift of the sensitivity coefficient of the FBG sensor, resulting in errors between the measured strain value and the true strain of the base, especially under complex curved surface or dynamic load conditions, the error is more significant. Therefore, in order to solve this error problem, the strain sensor needs to be calibrated in-situ.

[0003] The traditional in-situ calibration method mainly relies on static calibration in the laboratory, which is difficult to adapt to the demand of dynamic strain measurement under complex working conditions, and is not suitable for large-scale sensor calibration. In the prior art, the laser interference method can realize micron-level deformation measurement by inversely calculating the displacement field through interference fringes, but its essence is the detection of geometric displacement, which needs to indirectly deduce the strain value through spatial differentiation, resulting in the decrease of local strain sensitivity and error accumulation; the photogrammetry technology based on digital image correlation (DIC) can obtain full-field displacement data, but it needs to rely on artificial speckle marking and mathematical conversion of displacement data, which introduces additional errors, and the actual strain calculation accuracy is usually less than 50με, which is difficult to meet the high-precision calibration requirements. Although the existing calibration method selects a resistance strain gauge as a reference strain value, but this method needs to repeatedly paste wires and supply power, has weak anti-electromagnetic interference ability, and low measurement accuracy. In addition, the reference sensor and the sensor to be calibrated also have the problems of asynchronous data sampling and time misalignment. SUMMARY

[0004] In order to solve the problems of complex operation and low precision in the in-situ calibration of FBG sensors in the prior art, the present application provides an in-situ calibration system and method for component strain measurement based on a target facet grating.

[0005] One aspect of the present application provides a target facet grating-based in-situ calibration system for component strain measurement, comprising: a host computer, and an FBG sensor and a target facet grating arranged at the same position of a component to be detected for strain, the FBG sensor being a strain measurement device for the component, the FBG sensor being connected to an FBG demodulator, the FBG demodulator demodulating the center wavelength of the FBG sensor and transmitting the center wavelength of the FBG sensor to the host computer, the host computer being capable of calculating the center wavelength shift of the FBG sensor under different strain conditions of the component; the target facet grating being provided with a laser emitting device on one side and a light spot collecting device on the other side, the laser emitting device emitting laser at a preset angle into the target facet grating, the light spot collecting device collecting the diffraction light spot of the target facet grating on the focal plane and transmitting the image of the diffraction light spot of the target facet grating to the host computer; the host computer being capable of calculating the reference strain of the component according to the shift of the diffraction light spot of the target facet grating on the focal plane under different strain conditions of the component, so as to in-situ calibrate the strain value of the component corresponding to the center wavelength shift of the FBG sensor.

[0006] Further, the system further comprises a load loading device and an electric controller, the driving end of the electric controller being connected with the load loading device, the loading end of the load loading device being abutted with the component, and the signal output end of the host computer being connected to the electric controller, so as to drive the load loading device to generate different degrees of load to the component by controlling the electric controller, so as to make the component generate different strains.

[0007] Another aspect of the present application provides a calibration method of the target facet grating-based in-situ calibration system for component strain measurement as described above, the method comprising:

[0008] applying a load with a stepwise increase to a stepwise decrease to the component to be detected for strain, so as to make the component generate different degrees of strain over time;

[0009] synchronously collecting the center wavelength shift of the FBG sensor and the shift of the diffraction light spot of the target facet grating on the focal plane at each time, and calculating the reference strain of the component based on the shift of the diffraction light spot of the target facet grating on the focal plane; and forming a data set with the center wavelength shift of the FBG sensor and the reference strain of the component under the same load condition at each time;

[0010] sequentially inputting the center wavelength shift of the FBG sensor and the reference strain of the component under the same load condition at each time in the data set as a training pair into a preset mapping function model in time sequence, so as to train the mapping function model to obtain a mapping function between the center wavelength shift of the FBG sensor and the strain, and realize in-situ calibration of the FBG sensor.

[0011] Further, the offset amount of the target surface grating diffraction light spot on the focal plane is obtained by the following method:

[0012] From the target surface grating diffraction light spot images continuously collected at each moment, the centroid coordinates of the target surface grating diffraction light spot at each moment are calculated by using the gray weighted centroid method, and the pixel offset amount of the target surface grating diffraction light spot centroid at each moment relative to the initial state is calculated.

[0013] The pixel offset amount of the target surface grating diffraction light spot centroid is converted into a physical offset amount based on the preset pixel calibration parameters to determine the offset amount of the target surface grating diffraction light spot on the focal plane at each moment.

[0014] Further, the reference strain of the target surface grating diffraction light spot on the focal plane based on the offset amount calculation component includes: calculating the reference strain of the component based on a preset target surface grating strain calculation model, which is expressed as:

[0015]

[0016] In the formula, ε is the strain, Λ0 is the initial period of the target surface grating, m is the diffraction order, θ m is the mth-order diffraction angle, L is the vertical distance from the target surface grating to the focal plane, λ is the wavelength of the laser light source, and Δy is the offset amount of the target surface grating diffraction light spot on the focal plane.

[0017] Further, the mapping function model is realized by an MPGA-LSTM neural network model.

[0018] The mapping function model is trained to obtain the mapping function of the center wavelength offset of the FBG sensor and the strain, which includes:

[0019] Select the hyperparameters of the LSTM model as the optimization object, and divide the training set and the validation set in the data set;

[0020] Using the MPGA algorithm, for each set of hyperparameter combinations, the LSTM model is trained on the training set, and the prediction error of the LSTM model on the validation set after training is used as the fitness function to search for the optimal hyperparameter combination of the LSTM model.

[0021] Fixing the optimal hyperparameter combination searched by the MPGA algorithm, the LSTM model is trained on the training set by using the gradient descent algorithm to optimize the weight parameters inside the LSTM model, and finally the mapping function of the center wavelength offset of the FBG sensor and the strain based on the LSTM model is obtained.

[0022] Further, the LSTM model obtains the mapping function of the center wavelength offset of the FBG sensor and the strain based on the following steps:

[0023] The center wavelength shift of the FBG sensor at the current moment and the hidden state obtained at the previous moment are input into the forget gate of the LSTM model to obtain a forget gate control signal:

[0024] f t = σ(W f · [h t-1 , Δλ t ] + b f )

[0025] In the formula, f t is the forget gate control signal at the current moment, sigma is a sigmoid activation function, W f is the weight matrix of the forget gate, h t-1 is the hidden state obtained at the previous moment, Delta lambda t is the center wavelength shift of the FBG sensor at the current moment, b f is the bias term of the forget gate, and * represents matrix multiplication.

[0026] The center wavelength shift of the FBG sensor at the current moment and the hidden state obtained at the previous moment are input into the input gate of the LSTM model to obtain an input gate control signal and a candidate cell state:

[0027] i t = σ(W i · [h t-1 , Δλ t ] + b i )

[0028]

[0029] In the formula, i t is the input gate control signal at the current moment, W i is the control signal weight matrix of the input gate, b i is the control signal bias term of the input gate, is the candidate cell state, tanh is a hyperbolic tangent activation function, W c is the candidate cell weight matrix of the input gate, and b c is the candidate cell bias term of the input gate.

[0030] The cell state obtained at the previous moment and the candidate cell state obtained at the current moment are updated in coordination to obtain the cell state at the current moment:

[0031]

[0032] In the formula, C t is the cell state at the current moment, C t-1 is the cell state at the previous moment, and * represents element-wise multiplication.

[0033] The cell state at the current moment and the hidden state obtained at the previous moment are input into the output gate of the LSTM model to obtain the hidden state at the current moment:

[0034] o t =σ(W o ·[h t-1 ,Δλ t ]+b o )

[0035] h t =o t *tanh(c t )

[0036] In the formula, O t is the output gate control signal at the current moment, W o is the weight matrix of the output gate, b o is the bias term of the output gate, and h t is the hidden state at the current moment.

[0037] The strain value at the current moment corresponding to the center wavelength shift of the FBG sensor at the current moment is determined based on the hidden state at the current moment:

[0038] ε t =W y h t +b y

[0039] In the formula, ε t is the strain value at the current moment, W y is the weight matrix of the mapping function, and b y is the bias term of the mapping function.

[0040] The target facet grating-based component strain measurement in-situ calibration system and method provided by the application at least has the following advantages:

[0041] 1. By installing the target facet grating and the FBG sensor at the same position, using the diffraction effect of the periodic microstructure of the target facet grating on laser, and combining the light spot capturing device to capture the light spot displacement, a quantitative model of light spot displacement and target strain is established, so that dynamic strain in-situ calibration is realized.

[0042] 2. The target facet grating is designed with an ultrathin flexible substrate, has the characteristics of anti-electromagnetic interference and non-contact measurement, can be closely attached to a complex curved surface, has a sensitivity one order of magnitude higher than that of the FBG sensor, and does not need wiring or power supply, thereby simplifying the structure of the component strain measurement in-situ calibration system.

[0043] 3. An adaptive deep learning algorithm is introduced to make the model reflect complex influencing factors including installation stress and environmental interference, and finally realize intelligent compensation and in-situ calibration of the FBG measurement system. BRIEF DESCRIPTION OF DRAWINGS

[0044] Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The accompanying drawings are included to provide a description of the preferred embodiments and are not intended to limit the scope of the application. Moreover, like reference numerals are intended to represent like parts throughout all the drawings. In the drawings:

[0045] Figure 1 A schematic diagram of a target facet grating-based component strain measurement in-situ calibration system is provided for an embodiment of the present application;

[0046] Figure 2 A flowchart of a target facet grating-based component strain measurement in-situ calibration method is provided for an embodiment of the present application;

[0047] Figure 3 A test flowchart of a target facet grating-based component strain measurement in-situ calibration method is provided for an embodiment of the present application;

[0048] Figure 4 A load loading curve is provided for an embodiment of the present application;

[0049] Figure 5 A diffraction spot diagram of a target facet grating for a specific embodiment of the present application is provided;

[0050] Figure 6 A comparison diagram of strain before and after one level of diffraction spot displacement for a specific embodiment of the present application is provided;

[0051] Figure 7 A model structure diagram of an LSTM model for a specific embodiment of the present application is provided. DETAILED DESCRIPTION

[0052] To make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions of the embodiments of the present application will be described clearly and completely below with reference to the drawings of the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0053] In the actual measurement process of strain, due to the inconsistency of the parameters such as the thickness, length and installation angle of the adhesive layer during the laying and installation of the sensor, the sensitivity coefficient of the fiber Bragg grating sensor (hereinafter referred to as FBG sensor 4) pasted on the surface of the base body drifts, and there is an error between the measured strain and the real strain of the component. In view of this problem, the large component strain measurement in-situ calibration system and method based on the target surface grating 3 are proposed to calibrate the FBG sensor 4 strain measurement in-situ.

[0054] It should be noted that the target surface grating 3, also known as a reflective target surface grating 3 (Surface Target Grating, STG), is a kind of microstructure optical device, which produces diffraction effect on the incident laser through its periodic surface structure, and uses the displacement of the diffraction spot to measure the strain distribution of the target area. Among them, the target surface grating can be designed in the form of a thin film, with a base thickness of <100 μm, suitable for curved surface fitting, resistant to repeated bending, solving the problem of installation of traditional rigid sensors. Its flexible base and metal layer have high strain transmission efficiency, and can realize high-precision dynamic micro-strain measurement with a high-speed camera. Unlike traditional one-dimensional gratings, the surface grating has a two-dimensional periodic arrangement (such as square and hexagonal lattice) on the plane, which can independently modulate the x and y direction diffraction of the incident light. The target surface grating with calibrated sensitivity can be used for structural strain monitoring. Because the thin film target surface grating can realize high-precision strain measurement, using it as a reference standard can effectively improve the in-situ calibration accuracy.

[0055] Figure 1 The overall structure schematic diagram of the component strain measurement in-situ calibration system based on the target surface grating of an embodiment of the present application is shown, Figure 1 As shown in the figure, the component strain measurement in-situ calibration system based on the target surface grating 3 provided by the embodiment of the present application comprises an upper computer 7 and FBG sensors 4 and target surface gratings 3 arranged at the same position of the component. It should be noted that the component of the embodiment of the present application can be the aircraft wing 10 shown in the accompanying drawings, or other large components that need to detect strain. Figure 1

[0056] Among them, the FBG sensor 4 is a strain measurement device of the component, and the FBG sensor 4 is connected to the FBG demodulator 6 to demodulate the center wavelength of the FBG sensor 4 through the FBG demodulator 6, and transmit the center wavelength of the FBG sensor 4 to the upper computer 7. The upper computer 7 can calculate the center wavelength offset of the FBG sensor 4 based on the initial center wavelength of the FBG sensor 4. It should be noted that the strain of the component will cause the grating period of the FBG sensor 4 to change, which is manifested as the shift of the Bragg wavelength, and by establishing the conversion relationship between Δλ and strain (ε), the strain of the component can be finally detected. ​

[0057] Furthermore, a laser emitting device is provided on one side of the target surface grating 3, and a light spot acquisition device is provided on the other side. The laser emitting device emits a laser beam into the target surface grating 3 at a preset angle, and the light spot acquisition device collects the diffraction spot of the target surface grating 3 at the focal plane. It should be noted here that, in this embodiment of the invention, the imaging position of the diffraction spot on the light spot acquisition device is referred to as the focal plane of the light spot acquisition device, which can also be called the image plane or the target surface.

[0058] Furthermore, the host computer 7 receives the input signal from the light spot acquisition device, and calculates the offset of the diffraction spot of the target surface grating 3 on the focal plane under different strain conditions based on the initial position of the diffraction spot of the target surface grating 3. Based on the offset of the diffraction spot of the target surface grating 3 on the focal plane, the reference strain of the component is calculated to perform in-situ calibration of the component strain value corresponding to the center wavelength offset of the FBG sensor 4.

[0059] Specifically, such as Figure 1 The laser source 1 and collimator 2 on one side of the target grating 3 constitute a laser emitting device. The laser source 1 emits laser light of a preset wavelength, which is then collimated by the collimator 2 into an ideal, highly parallel collimated beam. The collimated laser beam illuminates the target grating 3 at a preset angle. The CCD camera 5 on the other side of the target grating 3 constitutes a spot acquisition device. The laser light diffracted by the target grating 3 is focused onto the target surface of the CCD camera 5 (i.e., the focal plane of the spot acquisition device). The CCD camera 5 acquires the diffracted spot. In this embodiment of the invention, the CCD camera 5 uses a high-sensitivity area array sensor with a high resolution of 5 million pixels (2592×1944), which can achieve a spot displacement detection accuracy of μm and a corresponding strain resolution better than 0.1με.

[0060] Furthermore, in this embodiment of the invention, the information input terminal of the host computer 7 is connected to the FBG demodulator 6 and the light spot acquisition device respectively, so as to calibrate the strain measurement of the FBG sensor 4 by acquiring the input signals of the FBG demodulator 6 and the light spot acquisition device, and determine the conversion relationship between the center wavelength offset of the FBG sensor 4 and the strain (ε).

[0061] Furthermore, the in-situ calibration system for component strain measurement based on the target surface grating 3 provided in this embodiment of the invention also includes a load loading device 9 and an electric controller 8. The loading end of the load loading device 9 abuts against the component, and applies a load to the component to induce corresponding strain. The driving end of the electric controller 8 is connected to the load loading device 9 to drive the load loading device 9. The electric controller 8 is controlled by a host computer 7 and can change the load applied to the component as needed.

[0062] In the specific embodiment of the present application, the electric controller 8 is a step motor, and the software of the host computer 7 controls the step motor to drive the load loading device 9, so as to realize the step loading and unloading cycle of the component, and then the central wavelength shift of the FBG sensor 4 under different strain conditions and the shift of the target surface grating 3 diffraction spot on the focal plane are collected, and then the FBG sensor 4 is calibrated in situ, and finally the conversion relationship between the central wavelength shift of the FBG sensor 4 and the strain (ε) is determined.

[0063] In order to facilitate the description of the method for in-situ calibration of the component strain measurement based on the target surface grating 3 provided by the embodiment of the present application, the corresponding relationship between the shift of the diffraction spot of the target surface grating 3 and the strain is simply discussed. It is assumed that the wavelength of the laser light source 1 is λ, the incidence angle is θ i , the period of the target surface grating 3 is Λ, and the sensitivity is S, wherein the diffraction equation is:

[0064] mλ=Λ(sinθ i +sinθ m ) (1)

[0065] In the above formula, m is the diffraction order, θ m is the mth order diffraction angle.

[0066] The period change of the target surface grating 3 caused by the strain is represented as:

[0067] ΔΛ=Λ-Λ0=Λ0ε (2)

[0068] In the above formula, ΔΛ is the period shift of the target surface grating 3, and Λ0 represents the initial period of the target surface grating 3.

[0069] The period change of the target surface grating 3 causes the diffraction angle shift:

[0070]

[0071] In the above formula, Δθ m is the mth order diffraction angle shift, θ m0 is the mth order initial diffraction angle, and θ m is the mth order diffraction angle after strain.

[0072] The simultaneous differentiation of both sides of equation (1) can be obtained:

[0073] 0=(sinθ i +sinθ m )ΔΛ+Λcosθ m Δθ m (4)

[0075] Substituting equations (1) and (2) into (3) gives

[0076]

[0077] Since ε<<1, (1+ε) 2 Approximately 1, further simplifying formula (5) can be obtained:

[0078]

[0079] Assume that the vertical distance from the plane where the target surface grating 3 is located to the focal plane of the CCD camera 5 is L, and the angle between the grating plane and the camera plane is Then the displacement Δy of the diffraction spot on the camera focal plane is:

[0080]

[0081] When the camera is parallel to the grating plane Under small angle, the relationship between Δy and the diffraction angle offset can be approximated as:

[0082] Δy≈Ltan(Δθ m )≈LΔθ m (8)

[0083] Substituting formula (8) into formula (6) to obtain the corresponding relationship between the spot displacement and the strain is:

[0084]

[0085] Formula (9) can be used as the strain calculation model of the target surface grating 3. In the actual calibration process, the displacement Δy of the diffraction spot on the camera focal plane is obtained, and then the reference strain of the component can be calculated.

[0086] Further, the sensitivity can be represented as:

[0087]

[0088] Therefore, by collecting the displacement of the diffraction spot of the target surface grating 3 on the focal plane of the CCD camera 5, the reference strain of the component can be calculated, and at the same time, the center wavelength offset of the FBG sensor 4 is collected, and the corresponding relationship between the center wavelength offset of the FBG sensor 4 and the strain of the component can be obtained.

[0089] The target surface grating 3 based component strain measurement in-situ calibration system provided by the embodiment of the present application utilizes the diffraction effect of the periodic microstructure of the target surface grating 3 on laser, combines with the CCD camera 5 to capture the displacement of the diffraction spot, establishes a quantitative model of the displacement of the diffraction spot and the target strain, so as to realize dynamic strain in-situ calibration. The target surface grating 3 adopts an ultrathin flexible substrate design, has the characteristics of anti-electromagnetic interference and non-contact measurement, can closely adhere to a complex curved surface, and has a sensitivity one order of magnitude higher than that of the FBG sensor 4, thereby significantly improving the calibration accuracy of the FBG sensor 4. In addition, the non-contact measurement characteristic of the target surface grating 3 avoids the additional error introduced by the traditional contact sensor.

[0090] Further, the target surface grating 3 adopts a flexible high polymer material substrate, has a small thickness, and can closely adhere to a complex curved surface, thereby solving the problem of difficult installation of the traditional rigid sensor on the curved component. Compared with the resistance strain gauge, the target surface grating 3 does not need wiring and power supply, and has strong anti-electromagnetic interference capability.

[0091] Further, Figure 2 and Figure 3 The specific flow of the target surface grating 3 based component strain measurement in-situ calibration method of the embodiment of the present application is shown, and the method will be described in detail with reference to Figure 2 It can be known that the method of the embodiment of the present application comprises the following steps:

[0092] S1, a load of a ladder type increment to a ladder type decrement is applied to a component to be detected for strain, so that the component generates different degrees of strain in a time sequence;

[0093] In the embodiment of the present application, the load of the ladder type increment to the ladder type decrement applied to the component to be detected for strain can be specifically as shown in Figure 4 The host computer 7 controls the electric controller 8 to drive the load loading device 9, realizes the ladder type loading and unloading cycle, and applies the ladder type load (0% to 80% yield strength, 5 levels of loading / unloading) to the component; 30 seconds are kept after each ladder loading, so as to ensure stable acquisition of the strain data; the electric controller 8 makes the equal strength beam rise and fall at a uniform speed of 0.1 mm / s, so as to avoid dynamic impact interference; and 3 complete loading-unloading cycles are performed.

[0094] S2, the center wavelength offset of the FBG sensor and the displacement of the target surface grating diffraction spot on the focal plane are synchronously acquired at each time, and the reference strain of the component is calculated based on the displacement of the target surface grating diffraction spot on the focal plane; and the center wavelength offset of the FBG sensor and the reference strain of the component obtained at each time under the same load condition form a data set;

[0095] In the embodiment of the present application, when the component generates strain, the input signals of the FBG demodulator 6 and the CCD camera 5 are collected by the host computer 7, and after data processing, the center wavelength shift of the FBG sensor 4 and the shift of the target surface grating 3 diffraction spot under the same load condition are determined.

[0096] S3, the center wavelength shift of the FBG sensor 4 and the reference strain of the component under the same load condition at each time in the data set are sequentially input into the preset mapping function model as a training pair to train the mapping function model to obtain the mapping function of the center wavelength shift of the FBG sensor 4 and the strain, and realize in-situ calibration of the FBG sensor 4.

[0097] In the embodiment of the present application, the center wavelength shift of the FBG sensor 4 under the same load condition is taken as the input of the mapping function model, and the reference strain of the component calculated according to the shift of the target surface grating 3 diffraction spot is taken as the output, the mapping function of the center wavelength shift of the FBG sensor 4 and the strain can be expressed as:

[0098] ε=f(Δλ) (11)

[0099] The embodiment of the present application discards the simplified model (ε=k·Δλ) of the traditional FBG sensor 4 that takes the center wavelength shift (Δλ) and the strain (ε) as a linear relationship. Through the preset mapping function model (ε=f(Δλ)), the present application firstly correlates and trains the Δλ of the FBG and the reference strain measured by the high-precision target surface grating 3; and then enables the model to deeply reflect the complex influencing factors including installation stress and environmental interference, and finally realizes intelligent compensation and in-situ calibration of the FBG measurement system, and converts it into a kind of high-precision in-situ strain sensor.

[0100] Reference Figure 3 In the embodiment of the present application, the center wavelength shift of the FBG sensor 4 and the shift of the target surface grating 3 diffraction spot on the focal plane are synchronously collected at each time, specifically, the center wavelength of the FBG sensor 4 is demodulated by the FBG demodulator 6, and the shift of the center wavelength of the FBG sensor 4 is calculated, the image collected by the CCD camera 5 is processed by the host computer 7 to obtain the spot image of the target surface grating 3 diffraction spot on the focal plane, the spot centroid of the diffraction spot is calculated based on the spot image, and then the shift of the spot centroid is calculated, and finally the reference strain value of the component for reference is obtained.

[0101] Further, the shift of the target surface grating 3 diffraction spot on the focal plane in step S2 is obtained by the following method:

[0102] S21, from the target surface grating 3 diffraction spot images collected continuously at each moment, the gray weighted centroid method is used to calculate the centroid coordinates of the target surface grating 3 diffraction spot at each moment, and the pixel offset of the target surface grating diffraction spot centroid relative to the initial state at each moment is calculated;

[0103] In the embodiment of the application, the gray weighted centroid method is represented as:

[0104]

[0105] In the above formula, (x c ,y c ) is the centroid coordinates of the target surface grating 3 diffraction spot, (x i ,y i ) is the pixel coordinates of the target surface grating 3 diffraction spot, I(x i ,y i ) is the gray value of the pixel point with coordinates (x i ,y i ), and n is the total number of pixels in the effective area of the target surface grating 3 diffraction spot.

[0106] S22, based on the preset pixel calibration parameters, the pixel offset of the target surface grating 3 diffraction spot centroid is converted into a physical offset, so as to determine the offset of the target surface grating 3 diffraction spot on the focal plane at each moment.

[0107] In the embodiment of the application, before calculating the centroid coordinates of the target surface grating diffraction spot at each moment, the spot image captured by the CCD camera 5 also needs to be preprocessed, specifically, the background noise of the spot image collected by the CCD camera 5 is deducted and Gaussian filtering is performed, so as to enhance the signal-to-noise ratio of the image.

[0108] Further, in the embodiment of the application, based on the characteristic that the spot energy is Gaussian distributed, the gray weighted centroid method is used to calculate the spot centroid coordinates by taking the pixel gray value as the weight, so as to accurately determine the centroid of the diffraction spot. Meanwhile, in the preferred embodiment of the application, the positioning accuracy can be improved to 0.1 pixel level by using the sub-pixel algorithm, and the subsequent calculation accuracy is further improved. The core advantage of this method is to abandon the edge detection idea which is easy to introduce errors, and instead make full use of the inherent energy distribution characteristics of the spot, so as to reduce the uncertainty of displacement measurement to microns.

[0109] Furthermore, the pixel calibration parameters represent the correspondence between pixels and actual distances. In step S22, converting the pixel offset of the centroid of the diffraction spot on the target surface grating 3 into a physical offset based on the preset pixel calibration parameters involves: placing a calibration plate with known precise scale (e.g., each grid is 100 micrometers) at the position of the grating target; taking an image of this calibration plate using a CCD camera 5; and counting how many pixels a certain number of grids occupy in the image. In practical applications, the actual distance of the centroid can be obtained based on the pixel displacement and the actual distance represented by each pixel.

[0110] For example, the actual physical length of 10 grids on the calibration plate is 100 μm / grid * 10 grids = 1000 μm. In the image, these 10 grids occupy 200 pixels. Therefore, the physical size (calibration coefficient) represented by each pixel is: k = physical size / number of pixels = 1000 μm / 200 pixels = 5 μm / pixel. This k is the key camera calibration parameter. For any measured pixel displacement Δy_pixels, a simple multiplication: Δy = Δy_pixels * k, yields the actual displacement of the centroid. It should be noted that, based on theory and actual measurements, the displacement of the centroid of the diffracted spot in the x-axis direction is negligible; therefore, the actual offset of the centroid of the target surface grating 3 diffracted spot before and after the load is applied is expressed as Δy.

[0111] Further, in this embodiment of the invention, calculating the reference strain of the component based on the offset of the diffraction spot of the target surface grating 3 on the focal plane includes: calculating the reference strain of the component based on a preset strain calculation model of the target surface grating 3, wherein the strain calculation model of the target surface grating 3 is:

[0112]

[0113] In the formula, ε is the strain, Λ0 is the initial period of the target grating 3, m is the diffraction order, and θ is the initial period of the target grating 3. m Δy is the m-th order diffraction angle, L is the vertical distance from the target grating 3 to the focal plane, λ is the wavelength of the laser source 1, and Δy is the actual offset of the centroid of the diffraction spot of the target grating 3.

[0114] Furthermore, in a specific embodiment of the present invention, the beam propagation path simulation of the optical path system of the target surface grating 3 in the YZ plane is as follows: Figure 5 As shown. A laser source is incident on the grating surface at an incident angle of 45°. After periodic microstructure diffraction, the first-order diffracted spot propagates to the target surface of the CCD camera at an angle of 4.25°. The geometric relationship between the incident light and the path of the diffracted spot verifies the diffraction equation (as shown in formula (1)), where Λ = 1 μm and m = 1. This figure visually reflects the diffraction angle shift Δθ caused by grating strain. mThe physical correlation with the spot displacement Ay (as in equation (8)) provides a geometric-optical basis for the calibration model.

[0115] Further, the intensity distribution of the first-order diffracted spot before and after the strain on the target surface of the CCD camera 5 is compared with the simulation as shown in Figure 6 Assuming that the incident collimated laser beam is a Gaussian beam of LG00 mode with a beam waist radius of 3 mm, when the strain ε = 10με, the beam offset angle is 0.001°, the spot centroid moves from (0, 0.371553 m) to (0, 0.3715855 m), and the spot displacement Ay = 32.5 μm in the negative direction of the y-axis. According to equation (10), the sensitivity S = 3.25 μm / με. High-resolution spot displacement detection (with a precision of μm level) verifies the micro-strain response characteristics of the target surface grating 3.

[0116] Further, referring to Figure 2 Before training the model, the preferred embodiment of the present application further includes data preprocessing and matching, that is, outlier rejection, filtering and denoising of the calculated reference strain values, and time series alignment of the two data by cross-correlation algorithm to eliminate timing error, so that the center wavelength shift of the FBG sensor 4 and the reference strain of the component under the same load condition are both values at the same time, to improve the accuracy of the model prediction.

[0117] Further, in a specific embodiment of the present application, the mapping function model is implemented by an MPGA-LSTM neural network model; the hyperparameters of the LSTM model are selected as the optimization object, and the training set and the validation set are divided in the data set; using the MPGA algorithm, for each set of hyperparameter combination, the LSTM model is trained on the training set, and the prediction error of the LSTM model on the validation set after training is used as the fitness function to search for the optimal hyperparameter combination of the LSTM model; fixing the optimal hyperparameter combination searched by the MPGA algorithm, the LSTM model is trained on the training set using the gradient descent algorithm to optimize the weight parameters inside the LSTM model, and finally the mapping function of the center wavelength shift of the FBG sensor and the strain based on the LSTM model is obtained.

[0118] Specifically, as Figure 7As shown, the LSTM network model in the embodiment of the present application is used to process the time sequence signal output by the FBG sensor 4, and the core advantage thereof lies in effectively capturing the time dependence in the strain measurement process through the gating mechanism. The network comprises an input layer, a hidden layer and an output layer. The LSTM model realizes selective memory and forgetting of historical information through three gating structures, i.e., a forgetting gate, an input gate and an output gate. The forgetting gate determines which information in the cell state at the previous moment needs to be retained; the input gate controls the importance of the current input information; and the output gate generates the output at the current moment according to the cell state. Such a mechanism enables the network to learn the response characteristics of the FBG sensor 4 under different load conditions, including the nonlinear drift caused by installation errors, temperature changes, material aging and other factors.

[0119] Specifically, the LSTM model obtains the mapping function of the center wavelength shift of the FBG sensor 4 and the strain based on the following steps:

[0120] The center wavelength shift of the FBG sensor 4 at the current moment and the hidden state obtained at the previous moment are input into the forgetting gate of the LSTM model to obtain a forgetting gate control signal:

[0121] f t =σ(W f ·[h t-1 ,Δλ t ]+b f ) (14)

[0122] In the formula, f t is the forgetting gate control signal at the current moment, σ is a sigmoid activation function, W f is a weight matrix of the forgetting gate, h t-1 is the hidden state obtained at the previous moment, Δλt is the center wavelength shift of the FBG sensor 4 at the current moment, b f is a bias term of the forgetting gate, and · represents matrix multiplication; in the embodiment of the present application, the forgetting gate can determine how much information is retained from the previous cell state, and thus can select to forget the historical noise irrelevant to the current strain measurement.

[0123] The center wavelength shift of the FBG sensor 4 at the current moment and the hidden state obtained at the previous moment are input into the input gate of the LSTM model to obtain an input gate control signal and a candidate cell state:

[0124] i t =σ(W i ·[h t-1 ,Δλ t ]+b i ) (15)

[0125]

[0126] where i t is the input gate signal at the current time, W i is the input gate control signal weight matrix, b i is the input gate control signal bias term, is the candidate cell state, tanh is the hyperbolic tangent activation function, W c is the candidate cell weight matrix of the input gate, b c is the candidate cell bias term of the input gate; in the embodiment of the present application, the input gate signal at the current time is analogous to a "filter" or a "write switch", and its value range is between [0, 1]. When the value is close to 1, it indicates that the new information is very important and needs to be remembered. When the value is close to 0, it indicates that the new information is not important and can be ignored. The candidate cell state is the potential new memory, and its value range is between [-1, 1]. Its role is to calculate the alternative content of the new information based on the center wavelength shift of the FBG sensor 4 at the current time and the hidden state at the previous time.

[0127] The cell state obtained at the previous time and the candidate cell state obtained at the current time are updated in coordination to obtain the cell state at the current time:

[0128]

[0129] where C t is the cell state at the current time, C t-1 is the cell state at the previous time, and * represents element-wise multiplication;

[0130] The cell state at the current time and the hidden state obtained at the previous time are input into the output gate of the LSTM model to obtain the hidden state at the current time:

[0131] o t = σ(W o · [h t-1 , Δλ t ] + b o ) (18)

[0132] h t = o t *tanh(c t ) (19)

[0133] where O t is the output gate control signal at the current time, W o is the weight matrix of the output gate, b o is the bias term of the output gate, h t is the hidden state at the current time;

[0134] Determine the strain value of the current moment corresponding to the center wavelength offset of the FBG sensor 4 of the previous moment based on the hidden state of the current moment:

[0135] ε t = W y h t +b y (20)

[0136] In the formula, ε t is the strain value of the current moment, W y is the mapping function weight matrix, and b y is the mapping function bias term.

[0137] Further, during the training of the conventional LSTM network, the weight parameters are usually randomly initialized, which is easy to fall into local optimization and affects the calibration accuracy. The present application introduces MPGA to globally optimize the hyperparameters of LSTM, preferably, the hyperparameters of LSTM specifically include: the number of hidden layer neurons: affecting the learning capacity and feature extraction ability of the network; the number of network layers: determining the fitting depth of the model to complex nonlinear relationships; learning rate: controlling the step size of parameter update, affecting the convergence speed and accuracy; dropout rate: preventing overfitting and improving the generalization ability of the model.

[0138] The particle position vector x i in the MPGA algorithm represents the set of optimizable hyperparameters of the LSTM network. The algorithm searches for the optimal initialization scheme through the particle swarm in the parameter space.

[0139] During the optimization process, the individual optimal position p i indicates the best parameter combination obtained by the i-th particle in the historical iteration process, which makes the root mean square error (RMSE) of the LSTM network on the training set reach the historical minimum of the particle. The global optimal position p g indicates the parameter combination with the best performance in the entire particle swarm, representing the best initialization scheme currently searched.

[0140] Particles update their positions according to individual experience and group experience:

[0141]

[0142] In the formula, v i is the particle velocity, x i is the particle position (i.e. the hyperparameters of LSTM), ω is the inertia weight, c1 and c2 are learning factors, and r1 and r2 are random numbers. The inertia weight ω gradually decreases during the iteration process, making the algorithm have strong global search ability in the early stage and turning to local fine search in the later stage.

[0143] Specifically, in the embodiment of the present application, the MPGA algorithm is used, for each set of hyperparameter combination, the LSTM model is trained on the training set, and the prediction error of the model on the validation set is used as the fitness function to search for the optimal hyperparameter combination of the LSTM model. Specifically: constructing the LSTM model structure according to the value configuration of the hyperparameters in the current iteration period; training the LSTM model using the training set data; calculating the root mean square error of the model prediction using the validation set data; taking the reciprocal or negative value of the root mean square error of the prediction as the particle fitness value; updating the particle velocity and position based on formulas (21) and (22); for each particle i, if the current fitness is better than the individual historical optimal fitness, update the position as p i ; for the entire population, if the fitness of a particle is better than the global historical optimal fitness, update the position as p g . Repeat the iteration process until the convergence condition is reached, which can be that the maximum number of iterations T is reached; or, the global optimal fitness is improved by less than a threshold value ε = 1e -6 for 10 consecutive generations; or the global optimal fitness reaches a preset target value. At this time, the global optimal position p g corresponds to the optimal hyperparameter combination of the LSTM model.

[0144] Further, in the embodiment of the present application, the gradient descent algorithm is used to train the LSTM model on the training set to optimize the weight parameters inside the model. Specifically: using the optimal hyperparameters to construct the LSTM network structure; using the gradient descent algorithm to optimize the model weight parameters on the training set; using the early stopping mechanism to prevent overfitting, and stopping training when the validation set error does not improve for 5 consecutive rounds; saving the trained LSTM model weight parameters. The LSTM model weight parameters can include weight matrices, mapping bias terms, and the above-mentioned hyperparameters, etc. The gradient descent algorithm is a conventional algorithm, which will not be described in detail.

[0145] Further, after the LSTM model is trained, the performance of the model can be verified by cross-validation and other methods, and the model parameters and algorithms can be optimized multiple times according to the feedback results of the verification to improve the calibration accuracy.

[0146] Further, the trained model is connected to the response output end of the FBG sensor 4 to realize in-situ calibration of the FBG sensor 4 for dynamic strain measurement of large structural parts, and to verify the strain measurement accuracy of the model.

[0147] The in-situ calibration system and method for component strain measurement based on target surface grating provided by the present application at least has the following advantages:

[0148] ​​​1. By installing the target surface grating 3 and the FBG sensor 4 at the same position, using the periodic microstructure of the target surface grating 3 to diffract the laser, combining the light spot collection device to capture the light spot displacement, a quantitative model of light spot displacement and target strain is established, so as to realize dynamic strain in-situ calibration.

[0149] 2. The target surface grating 3 adopts an ultra-thin flexible substrate design, has anti-electromagnetic interference and non-contact measurement characteristics, can closely fit complex curved surfaces, has a sensitivity one order of magnitude higher than the FBG sensor 4, and does not need wiring or power supply, simplifying the structure of the component strain measurement in-situ calibration system.

[0150] 3. The adaptive deep learning algorithm is introduced, so that the model can deeply reflect the complex influencing factors including installation stress and environmental interference, and finally realize intelligent compensation and in-situ calibration of the FBG measurement system.

[0151] In addition, those skilled in the art can understand that although some embodiments herein include certain features rather than others included in other embodiments, the combination of features of different embodiments means to be within the scope of the present application and forms different embodiments. For example, any one of the claimed embodiments can be used in any combination.

[0152] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A target facet grating based in-situ calibration system for component strain measurement, characterized by, The system comprises: an upper computer, an FBG sensor and a target surface grating arranged at the same position of a component to be detected for strain, the FBG sensor being a strain measurement device of the component, the FBG sensor being connected to an FBG demodulator, the FBG demodulator demodulating the center wavelength of the FBG sensor and transmitting the center wavelength of the FBG sensor to the upper computer, the upper computer being capable of calculating the center wavelength shift of the FBG sensor of the component under different strain conditions; the target surface grating is provided with a laser emitting device on one side and a light spot collecting device on the other side, the laser emitting device emits laser at a preset angle into the target surface grating, the light spot collecting device collects the diffraction light spot of the target surface grating on the focal plane and transmits the image of the diffraction light spot of the target surface grating to the upper computer; the upper computer can calculate the reference strain of the component according to the shift of the diffraction light spot of the target surface grating on the focal plane of the component under different strain conditions, so as to calibrate the strain value of the component corresponding to the center wavelength shift of the FBG sensor in situ.

2. The system of claim 1, wherein, The system further comprises a load loading device and an electric controller, the driving end of the electric controller being connected to the load loading device, the loading end of the load loading device being abutted to the component, and the signal output end of the upper computer being connected to the electric controller, so that the electric controller is controlled to drive the load loading device to generate different degrees of load on the component to make the component produce different strains.

3. A calibration method for a target facet grating based in-situ calibration system for component strain measurement according to claim 1 or 2, characterized in that, The method comprises: applying a load with a stepwise increase to a stepwise decrease to the component to be detected for strain, so that the component produces different degrees of strain over time; synchronously collecting the center wavelength shift of the FBG sensor and the shift of the diffraction light spot of the target surface grating on the focal plane at each time, and calculating the reference strain of the component based on the shift of the diffraction light spot of the target surface grating on the focal plane; and forming a data set of the center wavelength shift of the FBG sensor and the reference strain of the component under the same load condition at each time; sequentially inputting the center wavelength shift of the FBG sensor and the reference strain of the component under the same load condition at each time in the data set as a training pair into a preset mapping function model to train the mapping function model, so as to obtain the mapping function of the center wavelength shift of the FBG sensor and the strain, and realize in-situ calibration of the FBG sensor.

4. The method of claim 3, wherein, The shift of the diffraction light spot of the target surface grating on the focal plane is obtained by the following method: from the continuously collected diffraction light spot images of the target surface grating at each time, the gray weighted centroid method is used to calculate the centroid coordinates of the diffraction light spot of the target surface grating at each time, and the pixel shift of the centroid of the diffraction light spot of the target surface grating at each time relative to the initial state is calculated; based on the preset pixel calibration parameters, the pixel shift of the centroid of the diffraction light spot of the target surface grating is converted into a physical shift to determine the shift of the diffraction light spot of the target surface grating on the focal plane at each time.

5. The method of claim 4, wherein, The reference strain of the component based on the shift of the diffraction light spot of the target surface grating on the focal plane comprises: calculating the reference strain of the component based on a preset target surface grating strain calculation model, the target surface grating strain calculation model being represented as: where ε is the strain, Λ0 is the initial period of the target surface grating, m is the diffraction order, θm is the mth order diffraction angle, L is the vertical distance from the target surface grating to the focal plane, λ is the wavelength of the laser light source, and Δy is the offset of the target surface grating diffraction spot on the focal plane. m where ε is the strain, Λ0 is the initial period of the target surface grating, m is the diffraction order, θm is the mth order diffraction angle, L is the vertical distance from the target surface grating to the focal plane, λ is the wavelength of the laser light source, and Δy is the offset of the target surface grating diffraction spot on the focal plane.

6. The method of claim 5, wherein, The mapping function model is realized by an MPGA-LSTM neural network model; The mapping function model is trained to obtain a mapping function between the center wavelength shift of the FBG sensor and the strain, and the mapping function comprises: Selecting hyperparameters of the LSTM model as optimization objects, and dividing a training set and a validation set in the data set; Using an MPGA algorithm, for each set of hyperparameter combinations, training the LSTM model on the training set, and taking a prediction error of the LSTM model on the validation set after training as a fitness function to search for an optimal hyperparameter combination of the LSTM model; Fixing the optimal hyperparameter combination searched by the MPGA algorithm, using a gradient descent algorithm to train the LSTM model on the training set, optimizing weight parameters inside the LSTM model, and finally obtaining the mapping function between the center wavelength shift of the FBG sensor and the strain based on the LSTM model.

7. The method of claim 6, wherein, The LSTM model obtains the mapping function between the center wavelength shift of the FBG sensor and the strain based on the following steps: Inputting the center wavelength shift of the FBG sensor at a current moment and a hidden state obtained at a previous moment into a forget gate of the LSTM model to obtain a forget gate control signal: f t = σ(W f · [h t-1 , Δλ t ]+ b f ) wherein f t is the forget gate signal at the current time, σ is the sigmoid activation function, W f is the weight matrix of the forget gate, h t-1 is the hidden state obtained at the previous time, Δλt is the central wavelength shift of the FBG sensor at the current time, b f is the bias term of the forget gate, and · denotes matrix multiplication. Inputting the center wavelength shift of the FBG sensor at the current moment and the hidden state obtained at the previous moment into an input gate of the LSTM model to obtain an input gate control signal and a candidate cell state: i t = σ(W i · [h t-1 , Δλ t ]+ b i ) where i t is the input gate control signal at the current time, W i is the input gate control signal weight matrix, b i is the input gate control signal bias term, is the candidate cell state, tanh is the hyperbolic tangent activation function, W c is the input gate candidate cell weight matrix, b c is the input gate candidate cell bias term; Cooperatively updating the cell state obtained at the previous moment and the candidate cell state obtained at the current moment to obtain a cell state at the current moment: where C t is the cell state at the current time step, C t-1 is the cell state at the previous time step, and * denotes element-wise multiplication. Inputting the cell state at the current moment and the hidden state obtained at the previous moment into an output gate of the LSTM model to obtain a hidden state at the current moment: o t = σ(W o · [h t-1 , Δλ t ]+b o ) h t = o t tanh(c t ) In the formula, O t is the output gate control signal of the current moment, W o is the weight matrix of the output gate, b o is the bias term of the output gate, h t is the hidden state of the current moment; Determining a strain value at the current moment corresponding to the center wavelength shift of the FBG sensor at the current moment based on the hidden state at the current moment: e t = W y h t + b y In the formula, ε t is a strain value at a current time, W y is a mapping function weight matrix, and b y is a mapping function bias term.

Images