Online material recognition method based on optical fiber sensing

Abstract

The present invention relates to an online material identification method based on optical fiber sensing. First, an optical fiber grating demodulation module reads the central wavelength data collected by the optical fiber sensor in real time, and inputs the data stream of the central wavelength data to the receiving end on Labview, and Labview receives the central wavelength data in real time; the identified material slides on the upper surface of a double-layer silicone rubber encapsulated with an optical fiber sensor, and the central wavelength data of the optical fiber sensor during the sliding process is collected and saved; three characteristic values ​​of the wavelength data (mean maximum difference, range and standard deviation) are obtained through data processing, and updated and saved as a table type of Matlab, recorded as T; finally, T is input into a trained model for identification to obtain the material type, and Labview reads the identification result and displays it. The online material identification method based on optical fiber sensing of the present invention has the characteristics of high sensitivity, high accuracy, high stability and high speed.

Description

Technical Field

[0001] The invention belongs to the technical field of optical fiber intelligent sensing and relates to an online material identification method based on optical fiber sensing. Background Art

[0002] With the rapid development of science and technology, intelligent sensing robots are increasingly being used in production and quality inspection, greatly improving the efficiency and accuracy of industrial production. Among them, the tactile sensors responsible for sensing are devices that imitate human touch and can sense the movement, deformation, pressure and other information of the human body and the external environment. They have broad application prospects in the fields of smart medical care and intelligent robots. Most of the tactile sensors currently used use resistive, capacitive, piezoelectric, optical and photoelastic to transmit the sensed surface, but there are also certain limitations.

[0003] (1) Since most sensors are in direct contact, they will deform repeatedly in practice, resulting in reduced sensor performance;

[0004] (2) The variables that some tactile sensors need to sense during use include not only temperature changes but also external static and dynamic stimuli;

[0005] (3) The overall performance of sensitive materials is far lower than that of human skin;

[0006] (4) Problems with the sensing material itself, such as the hysteresis of photoelectric tactile sensors, the susceptibility of capacitive and piezoelectric tactile sensors to noise interference, and the dielectric properties being affected by temperature and distribution.

[0007] Existing material recognition technology perceives materials based on existing material properties and physical characteristics and then classifies and recognizes them. For example, patent application CN 114371210A proposes a beverage bottle material recognition method based on electrostatic detection technology, but it lacks the establishment of material characteristics and is easily disturbed by external electromagnetic interference. The literature (Research on Material Recognition Function of Fiber Bragg Grating Flexible Tactile Sensor, Journal of Scientific Instrumentation) uses the different thermal conductivity coefficients of different materials and the influence of temperature on the central wavelength of grating reflection to contact the grating with aluminum blocks, iron blocks and plastic blocks to obtain different wavelength drift rates, proving the manifestation of material differences in the central wavelength; however, the selected thermal conduction effect affects the recognition speed and process because the temperature change is a relatively slow process (taking more than 60s) and the contact object is stationary; it is difficult to distinguish when the contact objects have similar thermal conductivity coefficients by simply using the wavelength change diagram calibrated after the temperature change stabilizes.

[0008] Therefore, how to achieve high sensitivity, high accuracy, high stability and high speed in the process of tactile perception is a problem that material recognition technology needs to solve. Summary of the invention

[0009] The purpose of the present invention is to solve the above problems existing in the prior art and to provide a material online identification method based on optical fiber sensing.

[0010] To achieve the above object, the technical solution adopted by the present invention is as follows:

[0011] An online material identification method based on optical fiber sensing, firstly, makes the material to be identified slide on the upper surface of a double-layer silicone rubber encapsulated with an optical fiber sensor, and collects the central wavelength data of the optical fiber sensor during each sliding process, and then uses an online identification system to extract the characteristic value of the central wavelength data in real time, and analyzes it to obtain the material type.

[0012] The characteristic values ​​of the central wavelength data are the maximum difference of the mean of the central wavelength data (i.e. the difference between the maximum value of the central wavelength and the mean), the range and the standard deviation;

[0013] The online identification system includes a fiber Bragg grating demodulation module and an online identification platform. The online identification platform is Labview and Matlab. The fiber Bragg grating demodulation module reads the central wavelength data collected by the optical fiber sensor during each slip process in real time, and inputs the data stream of the central wavelength data to the receiving end of Labview (the data stream of the central wavelength data is divided into two groups, one group is input to a port on Labview and saves all wavelength data during the slip period in the form of a one-dimensional array, and the other group is input to the signal port of the "Statistics" function block of Labview). Labview receives the central wavelength data in real time and outputs the mean, maximum difference, range and standard deviation of the central wavelength data (Labview's existing output data has standard deviation std λB , maximum value max λB 、Minimum value min λB and average The system calculates the maximum mean difference, range and standard deviation based on the existing output data. The maximum mean difference is: The range is: max λB -min λB , standard deviation is: direct call), then save the mean maximum difference, range and standard deviation of the output central wavelength data to "wavelength characteristics.xlsx", Matlab reads "wavelength characteristics.xlsx", updates and saves it as table type, recorded as T, and finally inputs T into the trained model for identification to obtain the material type. The identification code is:

[0014] T.material=char(trainedModel.predictFcn(T));

[0015] The above code indicates that the trained model is used to predict and identify the three eigenvalues ​​in T, and the predicted results are saved in the material column of T in the form of characters (char); where trainedModel is the name of the trained model; trainedModel.predictFcn is the prediction and identification function of the model; trainedModel.predictFcn(T) indicates that T is the object to be predicted and identified, and the three eigenvalues ​​in T are read for prediction and identification;

[0016] The trained model is obtained by using the mean maximum difference, range and standard deviation of the central wavelength data corresponding to each material collected during the sliding process as input parameters and using the SVM algorithm in Matlab for classification training;

[0017] The corresponding relationship between the mean maximum difference, range and standard deviation of the material type and central wavelength data is as follows:

[0018] The maximum difference of the mean of material = coarse cloth is in the range of (0.00562~0.02648), the range of the extreme difference is in the range of (0.0328~0.0649), and the standard deviation is in the range of (0.005383~0.01618);

[0019] The maximum difference of the mean of material = PLA is in the range of (0.02260-0.09321), the range of the extreme difference is in the range of (0.0778-0.2092), and the standard deviation is in the range of (0.01554-0.05117);

[0020] Material = 800 grit sandpaper, the maximum difference of the mean is (0.02240~0.08545), the range of the extreme difference is (0.0473~0.1235), and the standard deviation is (0.008295~0.02800);

[0021] The units of the mean, maximum difference, range and standard deviation of the central wavelength data are all in nm.

[0022] As the preferred technical solution:

[0023] In the above-mentioned online material identification method based on optical fiber sensing, the sliding speed is 12 to 25 mm / s.

[0024] In the above-mentioned online material identification method based on optical fiber sensing, the optical fiber sensor is a fiber Bragg grating sensor (FBG) and is prestressed and stretched in advance to achieve slip perception.

[0025] In the above-mentioned online material identification method based on optical fiber sensing, the sliding process is completed by a sliding mechanism;

[0026] The sliding mechanism includes a sliding base, a prestressed stretcher and a transmission module; the sliding base is a base with a square groove, two layers of silicone rubber are poured in the square groove, and a fiber Bragg grating sensor is encapsulated between the two layers of silicone rubber; the pigtail of the fiber Bragg grating sensor is connected to the prestressed stretcher; the transmission module includes a slider, the slider is horizontally placed on the sliding base, and the identified material is attached to the lower surface of the slider.

[0027] In the above-mentioned online material identification method based on optical fiber sensing, the material of the sliding base is PLA.

[0028] In the above-mentioned online material identification method based on optical fiber sensing, the transmission module also includes a stepper motor, a ball screw, a driver, an MCU and a connector. The slider is connected to the ball screw through the connector so as to slide relative to the silicone rubber. The driver connects the MCU and the stepper motor to supply power to and control the rotation of the stepper motor. The MCU connects the driver and the display platform (i.e., a computer) to receive the start signal of the online identification platform and send a pulse signal to the driver. The stepper motor is connected to the ball screw through a screw slide.

[0029] The MCU sends pulses to the driver, and the driver controls the stepper motor to push the slider through the connecting parts to slide horizontally on the silicone rubber through the FBG to ensure the stability of the slider speed and constant force during sliding. The contact material is taken as the only variable, and finally the change process of the FBG center wavelength at the start and end of the start is recorded as the basis for material identification.

[0030] In the above-mentioned online material identification method based on optical fiber sensing, the stretching amplitude of the pigtail is controlled by an M6 bolt, and the pigtail is fixed by an M6 nut after being stretched by a prestressed stretcher, and the pigtail is connected to the bolt through a through hole.

[0031] The above-mentioned online material identification method based on optical fiber sensing is characterized in that the silicone rubber is 705 silicone rubber, which is automatically leveled and solidified by liquid pouring and has good sealing and linkage properties.

[0032] In the above-mentioned online material recognition method based on optical fiber sensing, the recognition accuracy of the trained model is above 95%.

[0033] The principle of the present invention is as follows:

[0034] The material identification method of the prior art uses the different thermal conductivity coefficients between different materials, which leads to the change in the center wavelength of the fiber Bragg grating temperature. It is necessary to conduct a heating experiment on the contact object. The selected thermal conduction effect is a relatively slow process (taking more than 60 seconds), and the contact object is static, which affects the recognition speed and process; it is difficult to distinguish when the thermal conductivity coefficients of the contact objects are similar by simply using the wavelength change diagram calibrated in the time period after the temperature change stabilizes. The present invention uses the different friction forces on the surfaces of different contact objects during the sliding process, which leads to the change in the center wavelength of the fiber Bragg grating strain, similar to the identification of the material of the contact object after a person touches the object with his fingers. Compared with classification and judgment based on experimental wavelength drift and calibration, the present invention can more accurately reflect the change of wavelength by extracting data features, and uses model algorithms for identification, so the data is more scientific and has a higher recognition rate for similar materials; and the results obtained by multiple experiments using the three characteristic values ​​of the present invention are better than the two-dimensional features of mean maximum difference and standard deviation and the four-dimensional features of maximum difference, range, standard deviation and mean square error used for model training; from the experimental conclusion, the classification and judgment using these three characteristic values ​​has a higher accuracy rate (the accuracy rates of the present invention for PLA, coarse cloth and 800 mesh sandpaper are all above 95%).

[0035] Beneficial effects:

[0036] The material online identification method based on optical fiber sensing of the present invention has the characteristics of high sensitivity, high accuracy, high stability and high speed.

[0037] High sensitivity: Fiber Bragg grating sensors have the following characteristics compared to piezoelectric and resistive sensors;

[0038] High accuracy: ① The conclusion of the trained model shows that it has a high accuracy rate; ② The stepper motor drives the ball screw to move horizontally, which is more stable than manual pushing and trolley pulling, preventing additional strain due to the traction itself; ③ Due to the wavelength changes caused by friction, stick-slip phenomenon, noise, etc. generated by the material, the three eigenvalues ​​of the present invention are more convenient and accurate in responding to the actual situation than manual comparison and calibration;

[0039] High stability: ① The running stability of the stepper motor ball screw kit itself; ② The single force on the base (horizontal movement force is vertically downward);

[0040] High speed: ① Compared with the use of material heat conduction to judge, the time required for slip detection is shorter. The speed of 12-25mm / s used in the present invention takes less than five seconds at a slip distance of 50mm; ② Identification is to call the training model, and there is no need for manual comparison with the calibration map. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is a structural schematic diagram of a fiber Bragg grating sensor;

[0042] Figure 2 It is a structural schematic diagram of the sliding mechanism;

[0043] Figure 3 It is a flow chart of the online material identification method based on optical fiber sensing;

[0044] Figure 4 is part of the data used for training;

[0045] Figure 5 and Figure 6 The scatter plot is formed after Matlab reads the data;

[0046] Figure 7 A schematic diagram of a model trained using the SVM algorithm;

[0047] Figure 8 Confusion matrix of the trained model;

[0048] Fig. 9 This is a schematic diagram of the Labview online recognition platform;

[0049] Fig.10 A flowchart for recording and saving central wavelength data and processing characteristic values ​​for Labview;

[0050] Fig.11 The central wavelength data during a sliding period is recorded and saved for Labview;

[0051] Fig.12 Provide examples of Matlab recognition codes and test results;

[0052] Fig.13 Identify before and after changes for the “wavelength signature” file;

[0053] Fig.14 This is the result of the material identification case;

[0054] Among them, 1 is a fiber Bragg sensor, 2 is a sliding mechanism, 3 is an online identification system, 1-1 is a fiber Bragg grating area, 1-2 is a pigtail, 1-3 is a fiber adapter, 2-1 is a sliding base, 2-2 is a slot, 2-3 is a prestressed tensioner, 2-4 is silicone rubber, 2-5 is a slider, 2-6 is a ball screw, and 3-1 is a fiber Bragg demodulation module. DETAILED DESCRIPTION

[0055] The present invention will be further described below in conjunction with specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and are not intended to limit the scope of the present invention. In addition, it should be understood that after reading the content taught by the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent forms fall within the scope limited by the appended claims of the application equally.

[0056] A material online identification method based on optical fiber sensing is completed by a sliding mechanism and an online identification system; Figures 1 to 3 As shown, the sliding mechanism 2 includes a sliding base 2-1, a prestressed tensioner 2-3 and a transmission module; the online identification system 3 includes a fiber Bragg grating demodulation module 3-1 and an online identification platform (Labview and Matlab); the sliding base 2-1 is a base with a square groove, and the material is PLA, and two layers of silicone rubber 2-4 (705 silicone rubber) are poured in the square groove, and a fiber Bragg grating sensor 1 is encapsulated between the two layers of silicone rubber 2-4; the transmission module includes a slider 2-5, a stepper motor, a ball screw 2-6, a driver, an MCU and a connector, the slider is horizontally placed on the sliding base, the lower surface of the slider is attached with the material to be identified, and the slider It is connected to the ball screw through a connector, the stepper motor is connected to the ball screw through a screw slide, and the driver is connected to the MCU and the stepper motor; the fiber Bragg grating tail fiber 1-2 is fixed by the prestressed stretcher 2-3 at the left and right slots 2-2, and then connected to the grating demodulation system 3-1 through the fiber adapter 1-3 to output the reflection spectrum (the tail fiber is stretched with an M6 bolt to control the stretching amplitude, and the tail fiber is fixed with an M6 nut after being stretched by the prestressed stretcher, and the tail fiber is connected to the bolt through a through hole), and the fiber Bragg grating demodulation module reads the central wavelength data collected by the fiber Bragg grating sensor during the sliding process in real time, and performs online material recognition through Labview;

[0057] First, a slider with the material to be identified attached to the lower surface is slid on the upper surface of a double-layer silicone rubber encapsulated with a fiber Bragg grating sensor, wherein the fiber Bragg grating sensor is prestressed with 2N and the sliding speed is 12-25mm / s. The central wavelength data of the fiber Bragg grating sensor during each sliding process is collected in real time, and then the characteristic data of the mean maximum difference, range and standard deviation of the central wavelength data are extracted in real time. Finally, the material type is obtained based on the characteristic data of the mean maximum difference, range and standard deviation of the central wavelength data;

[0058] Specifically, three materials were used: PLA, coarse cloth, and 800-mesh sandpaper. Each material was slid at a sliding speed of 25 mm / s, 15 mm / s, and 12 mm / s. The central wavelength data obtained for each sliding of the three materials were 33, 50, and 58, respectively. Twelve experiments were conducted at each speed, and more than a thousand wavelength data were obtained. Three eigenvalues ​​were extracted from the wavelength data of each sliding, totaling 36 sets of feature data for each of the three materials, with three eigenvalues ​​in each set; the data was imported into the svm vector machine algorithm of Matlab machine learning for classification training, and the optimal classification model was obtained, and the model was exported; when the eigenvalue of the central wavelength data was entered, Matlab would discriminate the material and output and save it, and then Labview software would read it and display it on the Labview online recognition platform.

[0059] The whole experimental process of the present invention is described in detail step by step below:

[0060] (1) More than 100 sets of data used for calibration ( Figure 4 The data of the identified materials (coarse cloth, PLA, 800-grit sandpaper) are imported into Excel through calibration test to obtain three characteristic values: the maximum difference of the mean (max-diff), the range (range) and the standard deviation (std). These three characteristic values ​​are obtained from the central wavelength data of the optical fiber sensor.

[0061] Figure 4 This is part of the more than 100 sets of data corresponding to the three materials in the implementation case under the above slip test. In each column, max-diff, range and std represent the characteristic values ​​of the central wavelength data obtained from a test, and the material is represented by "material".

[0062] The maximum difference of the mean of material = coarse cloth is in the range of (0.00562~0.02648), the range of the extreme difference is in the range of (0.0328~0.0649), and the standard deviation is in the range of (0.005383~0.01618);

[0063] The maximum difference of the mean of material = PLA is in the range of (0.02260-0.09321), the range of the extreme difference is in the range of (0.0778-0.2092), and the standard deviation is in the range of (0.01554-0.05117);

[0064] Material = 800 grit sandpaper, the maximum difference of the mean is (0.02240~0.08545), the range of the extreme difference is (0.0473~0.1235), and the standard deviation is (0.008295~0.02800);

[0065] (All data units are nm, and four significant figures are reserved when expressing in the present invention. Figure 4 The data listed in the table are accurate data calculated by Matlab and used for SVM model classification training.

[0066] (2) Matlab reads the data and imports it to form a scatter plot (the actual situation is three-dimensional, but due to Matlab's own reasons, only two eigenvalues ​​can be selected to correspond to xy to present a two-dimensional scatter plot), such as Figure 5 and 6 As shown;

[0067] (3) Figure 7 , 8 As shown, the SVM algorithm is used for classification training, and the model with the highest classification accuracy is finally selected (i.e., model 1.4; the overall classification accuracy of the model is 99.1%, the single-class PLA and coarse cloth are 100%, and the 800-grit sandpaper is 97.2%), and the training model is exported to the workspace for use as a tool for subsequent material identification;

[0068] (4) Start the experiment, open the Labview online recognition platform, click "Open Device", connect the demodulation module and MUC (i.e., single-chip microcomputer), and start reading the central wavelength data transmitted by the grating demodulation module in real time, such as Fig. 9 As shown;

[0069] (5) The sliding mechanism starts to control the slider to slide, Labview (online identification platform) starts to record and save data, and inputs the data stream of wavelength data to two receiving ends, one of which saves all wavelength data during the sliding period in the form of a one-dimensional array, and the other is input to the signal port of the "Statistics" function block of Labview. Click "Data Read", Labview starts to record and save wavelength data, and click "Start" at the same time, the sliding mechanism starts to control the slider to stop sliding. When the sliding displacement reaches 50mm, the motor stops rotating and data recording also stops. Fig.10 The flowchart for Labview records and saves the central wavelength data and processes the data to obtain the characteristic value, click Identify, and process the data to obtain the characteristic value);

[0070] (6) Save as Fig.11 Central wavelength data shown;

[0071] (7) The "Statistics" function block in Labview receives wavelength data in real time during the slip period. Its current output data has a standard deviation std λB , maximum value max λB 、Minimum value min λB and average Therefore, the online recognition system obtains three eigenvalues ​​based on the existing output data:

[0072] The maximum difference in means is:

[0073] The range is: max λB -min λB；

[0074] The standard deviations are: direct call;

[0075] The three eigenvalue data are used to display on the Labview (line recognition platform) interface;

[0076] At the same time, Matlab reads these three eigenvalues ​​and saves them to T (table type);

[0077] (8) Fig.12 As shown, the three feature values ​​are updated and saved to T (table type), and the trained model is substituted for recognition. The recognition code is: T.material = char (trainedModel.predictFcn (T)), and then T is saved to "wavelength characteristics.xslx", the code is: writetable (T, 'wavelength characteristics.xlsx'), as shown Fig.13 As shown;

[0078] like Fig.14 As shown, Labview's online recognition platform reads the data and displays the material type.

Claims

1. A material online identification method based on optical fiber sensing, characterized in that: First, the material to be identified is slid on the upper surface of the double-layer silicone rubber encapsulated with the optical fiber sensor, and the central wavelength data of the optical fiber sensor is collected during each sliding process. Then, the online identification system is used to extract the characteristic value of the central wavelength data in real time, and the material type is obtained through analysis. The characteristic values ​​of the central wavelength data are the maximum difference in the mean, the range and the standard deviation of the central wavelength data; the maximum difference in the mean refers to the difference between the maximum value of the central wavelength and the mean; The online identification system includes a fiber Bragg grating demodulation module and an online identification platform. The online identification platform is Labview and Matlab. The fiber Bragg grating demodulation module reads the central wavelength data collected by the optical fiber sensor in each sliding process in real time, and inputs the data stream of the central wavelength data to the receiving end of Labview. Labview receives the central wavelength data in real time and outputs the mean maximum difference, range and standard deviation of the central wavelength data. Then, the mean maximum difference, range and standard deviation of the output central wavelength data are updated and saved as Matlab's table type, recorded as T. Finally, T is input into the trained model to identify the material type. The identification code is: T.material=char(trainedModel.predictFcn(T)); The code indicates that the trained model is used to perform prediction and recognition based on the three feature values ​​in T, and the predicted results are saved in the material column of T in the form of characters; The trained model is obtained by using the mean maximum difference, range and standard deviation of the central wavelength data corresponding to each material collected during the sliding process as input parameters and using the SVM algorithm in Matlab for classification training; The corresponding relationship between the mean maximum difference, range and standard deviation of the material type and central wavelength data is as follows: The maximum difference of the mean of material = coarse cloth ranges from 0.00562 to 0.02648, the range is 0.0328 to 0.0649, and the standard deviation ranges from 0.005383 to 0.01618; The maximum difference of the mean of material = PLA ranges from 0.02260 to 0.09321, the range of the extreme difference ranges from 0.0778 to 0.2092, and the standard deviation ranges from 0.01554 to 0.05117; Material = 800 grit sandpaper, the maximum difference of the mean ranges from 0.02240 to 0.08545, the range of the extreme value ranges from 0.0473 to 0.1235, and the standard deviation ranges from 0.008295 to 0.02800; The units of the mean, maximum difference, range and standard deviation of the central wavelength data are all in nm.

2. According to claim 1, a material online identification method based on optical fiber sensing is characterized in that: The sliding speed is 12-25 mm / s.

3. The method for online material identification based on optical fiber sensing according to claim 1 is characterized in that: The optical fiber sensor is an optical fiber Bragg grating sensor and is pre-stressed and stretched.

4. The method for online material identification based on optical fiber sensing according to claim 3 is characterized in that: The sliding process is completed by a sliding mechanism; The sliding mechanism includes a sliding base, a prestressed stretcher and a transmission module; the sliding base is a base with a square groove, two layers of silicone rubber are poured in the square groove, and a fiber Bragg grating sensor is encapsulated between the two layers of silicone rubber; the pigtail of the fiber Bragg grating sensor is connected to the prestressed stretcher; the transmission module includes a slider, the slider is horizontally placed on the sliding base, and the identified material is attached to the lower surface of the slider.

5. The method for online material identification based on optical fiber sensing according to claim 4 is characterized in that: The material of the sliding base is PLA.

6. The method for online material identification based on optical fiber sensing according to claim 4 is characterized in that: The transmission module also includes a stepper motor, a ball screw, a driver, an MCU and a connector. The slider is connected to the ball screw through the connector. The driver connects the MCU and the stepper motor. The stepper motor and the ball screw are connected through a screw slide.

7. The method for online material identification based on optical fiber sensing according to claim 4 is characterized in that: The stretching amplitude of the pigtail is controlled by an M6 bolt, and after being stretched by a prestressed tensioner, the pigtail is fixed by an M6 nut, and the pigtail is connected to the bolt through a through hole.

8. The method for online material identification based on optical fiber sensing according to claim 4, characterized in that: The silicone rubber is 705 silicone rubber.

9. The method for online material identification based on optical fiber sensing according to claim 1, characterized in that: The recognition accuracy of the trained model is over 95%.

Images