Material identification method using fiber bragg grating (FBG) and material roughness

Abstract

The invention discloses a material identification method using a fiber bragg grating (FBG) and material roughness. According to the method, a three-axis sliding table, a fiber bragg grating (FBG), polydimethylsiloxane (PDMS) and a T-shaped rod are mainly used. The method comprises the following steps: firstly, packaging FBG in a PDMS material by using 502 glue, and controlling the packaging depth to be half of the thickness of a PDMS block; then, the packaged PDMS block is pasted on a T-shaped rod, and the T-shaped rod is fixed on a three-axis sliding table to slide; in the sliding process, FBG center wavelength change features are extracted and input into the LSTM neural network for classification and recognition. The device has the advantages of being small in size, high in sensitivity and resistant to electromagnetic interference, and is suitable for touch sensing of the fingertip of the intelligent robot.

Description

Technical Field

[0001] This invention belongs to the field of intelligent monitoring and control technology, specifically a material identification method utilizing the material roughness of fiber Bragg gratings (FBG). Background Technology

[0002] A tactile sensor is a device that can simulate the function of human skin, sense and measure physical contact information (such as pressure, vibration, texture, etc.). Its core task is to convert mechanical stimulation into a processable signal, which is a key technology for realizing intelligent machine interaction and biomimetic perception.

[0003] Currently, sensors used for roughness identification are mainly electromagnetic, piezoresistive, and capacitive. However, they are generally susceptible to electromagnetic interference, leading to signal distortion and often requiring complex shielding and filtering measures, which increases system cost and complexity. In contrast, fiber Bragg grating (FBG) sensors have advantages such as resistance to electromagnetic interference, ease of multiplexing, and miniaturization. Their working principle is based on the changes in grating period and refractive index caused by external physical quantities (such as pressure and temperature), which are ultimately reflected as a shift in the center wavelength.

[0004] Material identification often relies on electromagnetic signals derived from properties such as hardness and elasticity, or on analysis combining vision and neural networks; roughness features are rarely used. While FBG sensors offer high accuracy and strong anti-interference capabilities and are widely used in structural monitoring and medical sensing, their application in material identification remains limited. The main reason is that the friction generated when objects with different roughness slide can interfere with the center wavelength signal, significantly increasing the difficulty of identification. Summary of the Invention

[0005] In view of the shortcomings of the prior art, the purpose of this invention is to provide a material identification method that utilizes the material roughness of fiber Bragg gratings (FBGs) to solve the technical problems of existing tactile sensors being susceptible to electromagnetic interference, relying on complex signal processing circuits, and not making sufficient use of material roughness characteristics in material identification.

[0006] To achieve the above objectives, embodiments of the present invention disclose a material identification method utilizing the material roughness of a fiber Bragg grating (FBG), the method comprising the following steps:

[0007] Step S1: Encapsulate the FBG in a PDMS flexible material block to form a tactile sensing unit; slide the tactile sensing unit against the surface of the recognition material;

[0008] Step S2: Acquire and record the center wavelength signal of the FBG during the sliding process in real time, and perform wavelet transform denoising processing on the center wavelength signal.

[0009] Step S3: Extract preset statistical features from the center wavelength variation signal and construct a feature vector; the statistical features include: range, mean, interquartile range and mode;

[0010] Step S4: Input the feature vector into a pre-trained LSTM neural network model to output classification and recognition results.

[0011] Furthermore, the step of encapsulating the FBG in a PDMS flexible material block specifically includes:

[0012] PDMS flexible material liquid is poured into a mold and cured to form PDMS flexible material block;

[0013] The PDMS flexible material block is cut along its thickness direction to obtain two sub-blocks;

[0014] The FBG is placed between the two sub-blocks and fixed with adhesive, and the encapsulation depth of the FBG in the flexible material block is half the thickness of the flexible material block.

[0015] Furthermore, the tactile sensing unit slides against the surface of the recognition material in the following manner:

[0016] The tactile sensing unit is fixedly installed at one end of the T-shaped rod, and the other end of the T-shaped rod is fixedly installed at the moving end of the three-axis slide table; by controlling the three-axis slide table, the tactile sensing unit is driven to slide on the material surface at a set speed and direction.

[0017] Furthermore, the wavelet transform denoising process specifically includes:

[0018] Using wavelet transform basis functions, it can be expressed as:

[0019]

[0020] in, This represents the wavelet basis functions after scaling and translation; For wavelet basis scale, This is the translation amount; This indicates a signal indicating a change in center wavelength. Represents the mother wavelet function;

[0021] For the center wavelength change signal Multi-scale decomposition is performed to obtain wavelet coefficients at different scales and translational positions. These wavelet coefficients characterize the signal with respect to the center wavelength variation. and The degree of similarity between the upper and lower wavelet functions;

[0022] Select the set wavelet decomposition level and perform threshold processing on the decomposed wavelet coefficients;

[0023] The denoised center wavelength change signal is reconstructed based on the processed wavelet coefficients and used for subsequent feature extraction.

[0024] Furthermore, the FBG center wavelength change signal is caused by the friction force acting on the PDMS flexible material block during the sliding process to generate deformation, which is then transmitted to the packaged FBG to induce strain, and finally the strain causes the FBG center wavelength to shift.

[0025] Compared with the prior art, the beneficial effects of the present invention are:

[0026] (1) The present invention extracts roughness information by means of the dynamic change of the center wavelength of FBG caused by surface friction during the sliding process, which effectively overcomes the dependence of traditional tactile sensors on electromagnetic signals in material identification and significantly improves the robustness and applicability in complex electromagnetic environments.

[0027] (2) The present invention uses PDMS flexible material block to semi-buried encapsulate FBG and cooperates with T-shaped rod sliding structure, which not only effectively protects the grating from damage and contamination, but also converts the surface roughness of the material into a measurable wavelength offset signal. Combined with LSTM neural network to learn the timing features of the signal, while ensuring high recognition accuracy, the system hardware complexity and cost are greatly reduced.

[0028] (3) This invention constructs a multi-feature fusion and denoising process for roughness identification. It uses wavelet transform to denoise the center wavelength signal and extracts multi-dimensional features such as range, mean, interquartile range, and mode as network input. This method does not rely on precisely controlled sliding speed and can accurately identify various rigid and flexible materials under random speed conditions, demonstrating a wide operating range and strong engineering applicability. Attached Figure Description

[0029] Figure 1 This is a flowchart of the method of the present invention.

[0030] Figure 2 This is a schematic diagram of the FBG sensing principle when the slider slides in the method of the present invention.

[0031] Figure 3 This is a graph showing the FBG center wavelength response signal of different material surfaces at a sliding speed of 8 mm / s using the method of the present invention.

[0032] Figure 4 This is a graph showing the center wavelength response signal of the FBG on different material surfaces at a sliding speed of 10 mm / s, as per the method of this invention.

[0033] Figure 5This is a graph showing the center wavelength response signal of the FBG on different material surfaces at a sliding speed of 15 mm / s, as per the method of this invention.

[0034] Figure 6 This is a diagram illustrating signal denoising and center wavelength feature extraction using the method of the present invention. Detailed Implementation

[0035] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0036] Please see Figure 1 , Figure 2 , Figure 3 , Figure 4 , Figure 5 and Figure 6 This application provides a detailed description of the technical solutions provided in each embodiment.

[0037] like Figure 1 The flowchart of the method of the present invention shown includes the following specific steps:

[0038] FBG is encapsulated within a PDMS flexible material block to form a tactile sensing unit. The specific encapsulation process is as follows: A and B components of a two-component mixed curing adhesive are poured into a mold at a 1:1 ratio. The mold is 2.5cm long, 3cm wide, and 0.7cm high. After thorough mixing, the mixture is allowed to stand for 10 minutes until it completely cures and forms a PDMS block. Subsequently, the FBG is encapsulated and fixed using the aforementioned method to form a complete tactile sensing unit, which is then allowed to slide relative to the surface of the material to be identified.

[0039] Step S2: Acquire and record the center wavelength signal of the FBG during the sliding process in real time, and perform wavelet transform denoising processing on the center wavelength signal.

[0040] Step S3: Extract preset statistical features from the center wavelength variation signal and construct a feature vector; the statistical features include: range, mean, interquartile range and mode;

[0041] Step S4: Input the feature vector into a pre-trained LSTM neural network model to output classification and recognition results.

[0042] Using precision cutting tools, a horizontal cut was made at 0.35 cm along the height of the cured PDMS block, dividing the polydimethylsiloxane (PDMS) block into two flexible PDMS material blocks, each 2.5 cm long, 3 cm wide, and 0.35 cm high. The FBG was placed on the surface of one of the sub-blocks, ensuring the grating area was centered within it. 502 glue was evenly applied to the surface of the other sub-block as an adhesive. The two sub-blocks were precisely aligned and pressed together, completely encapsulating the FBG between the two PDMS sub-blocks. Crucially, the encapsulation depth of the FBG within the PDMS block was controlled to be half the total block thickness, i.e., 0.35 cm. This depth ensures adequate protection for the FBG while maintaining optimal responsiveness to strain caused by surface friction. The encapsulated PDMS-FBG composite was then glued to the end of a T-shaped rod, forming a complete tactile sensing unit.

[0043] The equipment used in this invention includes the following components: a three-axis slide stage, a Si-155 demodulator, a computer, a driver, a controller, and a 24V power supply. First, the driver and controller are connected to the three-axis slide stage. The controller allows setting the sliding direction and speed. The packaged FBG terminals are connected to the demodulator, which is then connected to the computer. During sliding, the demodulator converts the strain of the slide into a change in the FBG center wavelength, which is recorded in the Enlight software on the computer. The recorded center wavelength is used to calculate features using a MATLAB program, which are then fed into an LSTM neural network for training, ultimately yielding the classification prediction result.

[0044] like Figure 2 As shown, when a broadband light source propagates to a fiber Bragg grating, each small segment of the fiber, after its refractive index is changed, will only reflect a specific wavelength of light. This wavelength is called the Bragg wavelength, and its expression is:

[0045]

[0046] in, The center wavelength; Effective wavelength reflectivity; The grating period.

[0047] When the FBG receives an external load or a temperature change, and All of these will change, thus altering the size of the center wavelength. When affected by changes in temperature and stress, the change in center wavelength is expressed as:

[0048]

[0049] in, This is the offset of the center wavelength; This represents the effective elastic-optical coefficient of the optical fiber. This represents the amount of strain applied to the FBG; This represents the coefficient of thermal expansion of the optical fiber; This represents the thermo-optic coefficient of the optical fiber; It indicates the amount of change in ambient temperature.

[0050] During the sliding process of the PDMS block, due to the different material roughness and the coefficient of friction between the sliders... Different, according to the friction formula F represents the applied pressure, and the force on the slider can be calculated. According to the principles of mechanics of materials, the relationship between strain and stress in physics can be expressed as:

[0051]

[0052] in, For Young's modulus, For strain. When the FBG undergoes strain, according to the above formula, its refractive index and grating period will also change. This change will cause the center wavelength satisfying different Bragg conditions to shift, thus affecting the correspondence with different measured materials.

[0053] Figure 3 , Figure 4 and Figure 5 This study investigated the changes in the center wavelength of the FBG (Full-Fast Graph) when different materials slid along the x-axis at speeds of 8 mm / s, 10 mm / s, and 15 mm / s. The experimental materials used included 220-grit sandpaper, 600-grit sandpaper, aluminum, wood, glass, jute, flax, and silk. The aim was to accommodate both rigid and flexible materials while ensuring varying material roughness.

[0054] Figure 6 This study focuses on signal denoising and center wavelength feature extraction. Noise generation is unavoidable during the sliding process; therefore, denoising while preserving signal characteristics is crucial. Wavelet transform, as an improved method of Fourier transform, inherits and develops the localization concept of short-time Fourier transform while overcoming drawbacks such as the window size not changing with frequency. Wavelet transform provides a frequency-varying "time-frequency" window, making it an ideal tool for time-frequency signal analysis and processing.

[0055] Figure 6This paper demonstrates the denoising and feature extraction process of the center wavelength signal of the FBG (Fast-Fast Sensing). During the sliding sensing process, the original wavelength signal inevitably contains noise components due to environmental interference and inherent system noise. Therefore, this invention employs wavelet transform for signal preprocessing. This method overcomes the limitations of Fourier transform in non-stationary signal processing by adjusting the time-frequency window size, effectively suppressing noise while preserving the signal's abrupt change characteristics.

[0056] In this invention, the acquired raw signal is imported into the Matlab environment; the wden function in Matlab is used to decompose the wavelength signal. Wavelet transform is used for signal denoising; the specific steps are as follows:

[0057] Using wavelet transform basis functions, it can be expressed as:

[0058]

[0059] in, This represents the wavelet basis functions after scaling and translation; For wavelet basis scale, This is the translation amount; This indicates a signal indicating a change in center wavelength. Represents the mother wavelet function;

[0060] The decomposition layers were set to 3, 4, 5, and 6, and the optimal number of decomposition layers was determined by evaluating the signal-to-noise ratio (SNR) and mean square error (RMSE).

[0061] Thresholding is applied to the decomposed wavelet coefficients to eliminate noise components and then the signal is reconstructed.

[0062] Effective data segments are extracted from the denoised signal. In the experiment, four features are selected as input vectors to construct the training set: range, mean, interquartile range (IQR), and mode. The range and IQR reflect the discrete characteristics of the data, while the mean and mode reflect the central characteristics. For the range... The calculation formula is:

[0063]

[0064] mean The calculation formula is:

[0065]

[0066] Four-dimensional variance The calculation formula is:

[0067]

[0068] in, and The first and third quantiles are the values ​​of the first and third division points after the data is sorted in ascending order and divided into four equal parts.

[0069] For the mode The wavelength value that appears most frequently;

[0070] First, the center wavelength data of the FBG (Fast-Fast Gaussian Array) when 600-grit sandpaper slides at a speed of 15 mm / s was imported into Matlab. Using 5000 sampling points as one period, 3001-8000 sampling points were selected and defined as a vector. The sampling point is selected starting from 3001 because the noise is significant before the 3000th sampling point; then, these four feature values ​​are combined into a feature vector. .

[0071] The LSTM neural network architecture is adopted, which includes an input layer, two LSTM hidden layers, a fully connected layer, and an output layer.

[0072] The number of input nodes is set to 4, corresponding to the four statistical features;

[0073] The output layer uses the softmax activation function, and the number of output nodes corresponds to the number of material types.

[0074] Experimental data were collected from eight different materials: 220-grit sandpaper, 600-grit sandpaper, aluminum plate, cross-grained wood, glass, jute, flax, and silk.

[0075] For each material, 15 sets of data were collected at different speeds (8 mm / s, 10 mm / s, 15 mm / s), for a total of 120 training samples;

[0076] The dataset was divided into a training set and a test set in an 8:2 ratio.

[0077] Set the training parameters as follows: learning rate of 0.001, batch size of 16, and number of training epochs of 200.

[0078] Table 1 shows the material identification accuracy under random speeds. Material identification under constant speed sliding is a special case; therefore, testing the accuracy of material classification and identification under random speeds is generally applicable. A total of 120 random speed tests were conducted, resulting in 115 successful and correct classifications, achieving an accuracy rate of 95.5%.

[0079] Table 1. Material identification accuracy at random velocities

[0080] Material Number of accurate predictions / Total number of predictions accuracy 200 grit sandpaper 15 / 15 100% 600 grit sandpaper 15 / 15 100% Horizontal grain wood 14 / 15 93.3% aluminum plate 14 / 15 93.3% Glass 14 / 15 93.3% jute 14 / 15 93.3% flax 14 / 15 93.3% silk 15 / 15 100% total 115 / 15 95.5%

[0081] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A material identification method utilizing fiber Bragg gratings (FBGs) and material roughness, characterized in that, The method steps include: Step S1: Encapsulate the FBG in a PDMS flexible material block to form a tactile sensing unit; slide the tactile sensing unit against the surface of the recognition material; Step S2: Acquire and record the center wavelength signal of the FBG during the sliding process in real time, and perform wavelet transform denoising processing on the center wavelength signal. Step S3: Extract preset statistical features from the center wavelength variation signal and construct a feature vector; the statistical features include: range, mean, interquartile range and mode; Step S4: Input the feature vector into a pre-trained LSTM neural network model to output classification and recognition results.

2. The material identification method using fiber Bragg gratings (FBG) and material roughness according to claim 1, characterized in that, The step of encapsulating FBG in a PDMS flexible material block specifically includes: PDMS flexible material liquid is poured into a mold and cured to form PDMS flexible material block; The PDMS flexible material block is cut along its thickness direction to obtain two sub-blocks; The FBG is placed between the two sub-blocks and fixed with adhesive, and the encapsulation depth of the FBG in the PDMS flexible material block is half the thickness of the PDMS flexible material block.

3. The material identification method using fiber Bragg gratings (FBG) and material roughness according to claim 1, characterized in that, The tactile sensing unit slides against the surface of the recognition material in the following way: The tactile sensing unit is fixedly installed at one end of the T-shaped rod, and the other end of the T-shaped rod is fixedly installed at the moving end of the three-axis slide table; by controlling the three-axis slide table, the tactile sensing unit is driven to slide on the material surface at a set speed and direction.

4. The material identification method using fiber Bragg gratings (FBG) and material roughness according to claim 1, characterized in that, The wavelet transform denoising process specifically includes: Using wavelet transform basis functions, it can be expressed as: in, This represents the wavelet basis functions after scaling and translation; For wavelet basis scale, This is the translation amount; This indicates a signal indicating a change in center wavelength. Represents the mother wavelet function; For the center wavelength change signal Multi-scale decomposition is performed to obtain wavelet coefficients at different scales and translational positions. These wavelet coefficients characterize the signal with respect to the center wavelength variation. and The degree of similarity between the upper and lower wavelet functions; Select the set wavelet decomposition level and perform threshold processing on the decomposed wavelet coefficients; The denoised center wavelength change signal is reconstructed based on the processed wavelet coefficients and used for subsequent feature extraction.

5. The material identification method using fiber Bragg gratings (FBG) and material roughness according to claim 1, characterized in that, The FBG center wavelength change signal is caused by the friction force acting on the PDMS flexible material block during the sliding process, which causes deformation. The deformation is then transmitted to the packaged FBG to induce strain, and finally the strain causes the FBG center wavelength to shift.

Images