CN116147811A - Optical Microcavity Based Microforce Sensing Device Using Machine Learning - Google Patents

Abstract

The invention discloses a micro-force sensing device based on an optical microcavity by utilizing machine learning, which comprises the steps that firstly, a laser generates a continuous sweep frequency laser signal, and the laser signal is coupled into the optical microcavity after the polarization state is adjusted by a polarization controller; because the optical microcavity has an ultrahigh quality factor, the coupled optical signals form a multi-resonant-mode spectrum with a high Q value; collecting a large number of spectrograms under the same measuring force condition, inputting the spectrograms into a machine learning model for training and testing, and establishing a neural network model capable of identifying the spectrum similarity of the resonant mode; and collecting a multi-resonance-mode spectrum under the action of the force to be measured, and leading the spectrum into a neural network model after the training is successful, so as to output the size of the force to be measured. The problem that tracking cannot be performed when a single mode moves out of a scanning range is solved, the applicability of the micro force sensor is improved, compared with manual calculation, manpower and material resources are saved, and meanwhile the measuring speed and accuracy are improved. At the same time, the micron-scale device size facilitates packaging and integration.

Description

Optical Microcavity Based Microforce Sensing Device Using Machine Learning

technical field

The invention belongs to the technical field of optical sensing, and in particular relates to an optical microcavity micro force sensing device using a machine learning algorithm.

Background technique

Micro force sensing and measurement has always been an important frontier topic in the field of micro-nano electromechanical technology development. Micro force sensing systems with high sensitivity, high precision, and large dynamic measurement range are widely used in aerospace, medical, manufacturing and other fields. prospect. In almost all industries, products must be more compact on the basis of higher performance, which has higher requirements for the detection structure size and integration of micro force sensors. However, traditional micro force sensors such as piezoelectric and capacitive sensors have defects such as insufficient resolution, easy to be disturbed by the outside world, and large size, which bring a lot of inconvenience to the measurement of micro force.

In order to improve the resolution of micro force sensing and solve the deficiencies in traditional detection technology, many new methods and structures have been proposed in recent years. Utilize the high-temperature superconducting suspension technology to build a micro-force measuring device, which can effectively isolate various interferences, such as CN 102853954 B; based on the bubble microcavity fine-core fiber, make it produce a Fabry-Perot microcavity in the fiber, thereby making A temperature-insensitive, bending-insensitive bubble microcavity fine-core optical fiber tension sensor has good application prospects, such as CN 112880888 A; the effective refractive index modulation of the waveguide and the resonant wavelength frequency shift produced by the Bragg reflector are used. The detection provides pressure sensing, which can realize the measurement of the pressure value and the small change amount of the liquid and gas environment, the manufacturing process is simple, and the size is small, such as CN 105424261 B.

Among the many micro force measurement methods, the whispering gallery mode optical microcavity has the characteristics of extremely high quality factor and the resonance frequency is extremely sensitive to the shape, especially the microbubble cavity with a hollow structure or the polymer microcavity with a small elastic modulus. The cavity can obtain higher sensitivity and lower detection limit than traditional micro-nano optical sensors, and has a good application prospect in the field of micro-force sensors. The shape and refractive index of the microcavity are changed by the tiny force, so that the resonant frequency of the output light has a certain frequency shift, and the value of the measured force can be obtained by monitoring the magnitude of the frequency shift. This sensing mechanism has the advantages of high sensitivity and high precision, and the whispering gallery mode microcavity coupling system has a compact structure, which is conducive to the miniaturization and integration of micro force sensors. However, there are still some problems in the existing micro-force sensors based on optical microcavities. First, the actual force cannot be directly extracted from the output resonance spectrum. The force sensing is realized by monitoring the relative movement of the resonance mode relative to its initial resonance frequency, so only the relative value of the force can be measured; secondly, due to the sensing method Based on the tracking of the resonant frequency of the same mode, its dynamic range is limited by the scanning range of the pump laser. When the measurement range increases, this resonant mode may drift out of the scanning range, resulting in limited dynamic measurement range of the sensor.

Contents of the invention

Aiming at the problems existing in the prior art, the object of the present invention is to provide a micro force sensing method and device based on an optical microcavity using a machine learning model. The sensing method and device are based on the spectral pattern formed by multiple whispering gallery mode resonance peaks of the optical microcavity, by using deep learning and/or other machine training algorithms to form a characteristic spectral pattern library for a large number of measurement results of different measurement forces, and obtain Large dynamic range, real-time, high-resolution micro-force measurement results, and has the advantages of small size, light weight, and easy integration.

In order to solve the problems of the technologies described above, the technical solution adopted in the present invention is:

A micro-force sensing device based on optical microcavity using machine learning, including a microcavity microforce sensing and demodulation system, the microcavity microforce sensing and demodulation system consists of a laser, a polarization controller, a coupling waveguide, an optical microcavity, Composed of photodetectors, high-speed oscilloscopes and data processing and machine learning modules;

Among them, the laser generates continuous tunable laser light, the continuous tunable laser light is used as the input light, the polarization controller controls the polarization state of the input light, and the coupling waveguide couples the light into the optical microcavity; the optical microcavity is used as a high-sensitivity force sensor device to form a A series of high-Q multi-resonant modes; the optical microcavity is equipped with a piezoelectric fine adjustment module and a clamping coarse adjustment module. The piezoelectric fine adjustment module is used to load the micro force to be measured, and the clamping coarse adjustment module is used to fix the optical microcavity. and coarsely adjust the position of the microcavity; the photodetector is used to receive the optical signal from the optical microcavity coupling system, and convert it into an electrical signal and input it to a high-speed oscilloscope, which is used to monitor the resonant mode spectrum of the microcavity; data processing and machine learning The module is used to collect and process pattern spectrum data, and perform machine learning and analysis processing on pattern spectrum data. Further, the laser is a continuous wavelength tunable laser, which is used in conjunction with an external signal generator to inject frequency-swept narrow-band laser signals into the system.

Further, the coupling waveguide and the optical microcavity form an Add-pass coupling structure or an Add-drop coupling structure, and the coupling waveguide can be a tapered optical fiber, a coupling prism or an integrated waveguide.

Further, the material of the optical microcavity can be quartz SiO ₂ , polymer material such as PDMS, PMMA, or crystal material such as CaF ₂ , BaF ₂ or MgF ₂ .

Further, the optical microcavity adopts a whispering gallery mode optical microcavity, which can be solid or hollow, and its shape can be spherical, bottle-shaped, cylindrical or ellipsoidal; according to its mode structure, it can form a multi-resonance mode spectrum.

Further, the piezoelectric fine-tuning module adopts a piezoelectric ceramic stacking method, and by changing the voltage of the PZT, an adjustable axial force can be applied to the microcavity; under the action of the driving force, the microcavity multi-resonance peak mode spectrum generates frequency shift.

Further, the data processing and machine learning module processes the collected multi-resonant peak mode spectral data into a normalized two-dimensional optical barcode, which includes the spectral interval, linewidth, resonance wavelength and coupling of the multi-resonant mode. depth and other information.

Further, the optical barcode is composed of many rectangular areas, each rectangular area is a "bit" of the barcode, the width of each rectangular area represents the line width of the mode, and the color reflects its coupling depth, thereby generating a Uniquely corresponds to an optical barcode.

Further, the data processing and machine learning module collects multi-resonant peak mode spectral data under different conditions (micro force size and stretching length), and processes to obtain standard two-dimensional optical barcodes corresponding to different forces; The barcode is used as input, and the micro-force and stretching length under the corresponding conditions are used as output. After training with neural network algorithm, a convolutional neural network regression model that can predict the input data versus stress is obtained; the force to be measured is applied to the optical microcavity The coupling system obtains the optical barcode to be tested after spectral collection and data processing, and inputs the two-dimensional optical barcode image into the trained neural network model for regression analysis, obtains the regression value of the force to be measured, and obtains the magnitude of the absolute force.

Further, the neural network model is characterized in that: the functions of the neural network model can be realized by most convolutional neural network technologies, including but not limited to VGG19 neural network, GoogLeNet neural network and the like.

The working principle of the above-mentioned micro force sensor based on optical microcavity using machine learning is as follows: firstly, the laser generates a continuous frequency-sweeping laser signal, which is coupled into the optical microcavity after the polarization state is adjusted by the polarization controller; because the optical microcavity has a super high quality factor , the coupled output optical signal forms a high-Q multi-resonance mode spectrum; collect a large number of spectrograms under the same measurement force and input them into the machine learning model for training and testing, and establish a neural network model that can identify the similarity of the resonant mode spectrum; Collect the multi-resonance mode spectrum under the action of the force to be measured, import the neural network model after the training is successful, and output the magnitude of the force to be measured; it is worth noting that this sensing method is based on the feature recognition of the multi-mode resonance spectrum without tracking A single resonance mode, using machine learning algorithms, can achieve large dynamic measurement range and real-time absolute micro force sensing.

The present invention has the following advantages:

(1) The present invention provides a kind of identification and matching method based on optical microcavity multi-mode resonant spectrogram, analyzes the overall spectral feature quantity in the mode scanning range, overcomes the problem of the single-mode tracking method measurement range limitation, utilizes simultaneously The characteristics of the high Q value of the whispering gallery mode can realize the measurement of high sensitivity and large dynamic range.

(2) The present invention realizes the measurement of absolute force by using machine learning algorithm regression analysis, solves the problem that the traditional microcavity force sensor can only measure the relative value of force, and has the characteristics of real-time, fast and high precision.

(3) The sensing device involved in the present invention adopts a whispering gallery mode microcavity coupling system, which can further improve sensing sensitivity and resolution by selecting materials and designing a hollow structure. The microcavity system is small in size and light in weight, which is conducive to miniaturization and Integration provides a new method for the application of force sensors in semiconductor, electronics, aerospace and medical fields.

Description of drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings that need to be used in the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those skilled in the art, other drawings can also be obtained based on these drawings without creative effort.

Fig. 1 is the structure diagram of micro force sensor;

Figure 2 is the multi-resonance mode transmission spectrum diagram under the action of different forces;

Fig. 3 is the two-dimensional multi-resonance mode spectrum pattern formed after data processing of the multi-resonance mode spectrum;

Figure 4 is a method for micro force sensing using machine learning;

Fig. 5 is the training set and test set in the present embodiment;

Figure 6 is an optical barcode corresponding to different force sizes;

Figure 7 is a comparison chart between the predicted value and the actual value;

Fig. 8 is the fitting curve and the real value curve of the predicted value of the tiny force output by the regression model;

Fig. 9 is the error histogram of regression model;

Figure 10 shows the regression prediction results of optical barcodes corresponding to 10 pull forces of different sizes.

Labels in the figure: 1-Laser, 2-Polarization controller, 3-Coupling waveguide, 4-Optical microcavity, 5-Piezoelectric fine adjustment module, 6-Clamping coarse adjustment module, 7-Photodetector, 8-High speed Oscilloscope, 9-data processing and machine learning module.

Detailed ways

In order to make the technical means, creative features, goals and effects achieved by the present invention easy to understand, the present invention will be further described below in conjunction with specific embodiments.

Fig. 1 is a schematic structural diagram of a micro-force sensor based on an optical microcavity using machine learning proposed by the present invention. The micro force sensor consists of a laser 1, a polarization controller 2, a coupling waveguide 3, an optical microcavity 4, a piezoelectric fine adjustment module 5, a clamping coarse adjustment module 6, a photodetector 7, a high-speed oscilloscope 8 and data processing and machine learning The module 9 forms a micro-cavity micro-force sensing and demodulation system, which is the main structure of the micro-force sensor; wherein, the laser 1 is connected to the input end of the coupling waveguide 3 after passing through the polarization controller 2. In order to generate frequency-swept laser signals, the polarization controller controls the polarization state of the input laser, and the coupling waveguide 3 couples the laser light into the optical microcavity 4; Value multi-resonance mode spectrum; the clamping coarse adjustment module 6 is used to fix the optical microcavity and coarsely adjust the axial position of the microcavity, the piezoelectric fine adjustment module 5 is used to load the micro force to be measured; the photodetector 7 is used to receive the optical The optical signal of the microcavity coupling system is converted into an electrical signal and input to a high-speed oscilloscope. The high-speed oscilloscope 8 is used to monitor the resonance mode spectrum of the microcavity; the data processing and machine learning module 9 collects and processes the multi-resonance mode spectrum data, and The mode spectrum data is processed by machine learning and analysis, and finally the measured micro force results are obtained.

Further, in this embodiment, the optical microcavity is a _SiO2 bottle-shaped microcavity, and its preparation adopts the arc discharge method of an optical fiber fusion splicer. Its axial profile is approximately parabolic, and the axial length of the convex part is about 443 μm. The maximum diameter of the cavity is about 181 μm, and the minimum diameter is about 125 μm. The coupling waveguide adopts a tapered optical fiber with a tapered waist diameter of about 2 μm, which is prepared by the "thermal drawing method".

Fig. 2 is the multi-resonance mode transmission spectra under different forces. On the basis of the optical path structure in Figure 1, firstly, the appropriate microcavity transmission spectrum signal is obtained through the adjustment of the clamping coarse adjustment module, as shown in the lowermost spectral line in Figure 2; The gradually increasing voltage of the interval value applies to the microcavity the small force (in the elastic range of the microcavity material) that is gradually increased along the axial equidistant intervals. In the embodiment, _SiO2 material is used, which can be obtained by mechanical formula or finite element calculation, When the tensile force increases by 0.18N, the corresponding length changes by about 2 μm; with the increase of the tensile force value, the overall frequency shift of the multi-resonance mode spectrum makes each force magnitude correspond to a characteristic spectrum, and microcavities with different materials, sizes and mode field distributions will produce different results.

Figure 3 is a two-dimensional multi-resonance mode spectrum pattern formed after data processing of the multi-resonance mode spectrum. Each line represents a mode in the spectrum, the line width indicates the line width of the mode, and the color reflects its coupling depth. The optical barcode patterns corresponding to different measurement forces have different mode positions, mode numbers, line widths, and coupling depths, so the characteristics of the spectral patterns corresponding to each measurement force are unique, which is equivalent to the two-dimensional optical barcode of this measurement force.

Figure 4 is an implementation flowchart of a method for micro force sensing using machine learning, using a convolutional neural network (CNN) based on VGG19 image regression. Considering the influence of temperature drift and air flow on the frequency shift of the spectrum, the multi-resonance mode spectrum under the same force is first sampled multiple times (50 groups), and a large amount of sampled spectral data is processed into a two-dimensional Resonant mode spectral pattern, and used as a training set to build a neural network model that can perform regression calculations; based on the above-established neural network regression model, import the spectral pattern to be measured into the neural network, and the corresponding absolute force of the spectral pattern will be output .

The functions of the neural network model can be realized by most convolutional neural network technologies, such as VGG19 network, GoogLeNet network and the like. The network depth can improve the accuracy of the network to a certain extent, and the VGG19 network is used in this embodiment. In order to reduce the number of network training parameters, the entire convolutional network uses convolutions of size 3×3. In this embodiment, the imageInput layer is matched with the input data format, the number of neurons in the output layer is matched with the number of output forces, a dropout layer is added to the convolutional neural network model to prevent overfitting, and the output is output through the regression layer without adding softmax layer. Finally, adjust the parameters of each module to optimize the network effect.

Fig. 5 is the training set and test set in this embodiment. In the process of model training, all the optical barcodes of the samples are used as input, and the micro force and stretching length under the corresponding conditions are used as output. Supervisedly train the neural network model according to the output vector, and continuously adjust the weight of each layer of the network to realize the multi-mode feature information contained in the optical barcode, such as line width, coupling depth, etc., to the feature mapping of the output vector, complete mode of training. After the training is completed and the model evaluation is reasonable, when the optical barcode drawn by importing the test set data is shown in Figure 6, the system can predict the corresponding absolute force according to the characteristic information contained in the barcode, and the output results are shown in the table 1.

In the present invention, a large number of sample data collected in previous experiments are all used as training sets and standard databases. On the one hand, the number of training samples has a direct and important impact on the accuracy of the trained model, so the more samples, the better the training effect. On the other hand, the accuracy of the micro-force interval size corresponding to the standard database is also closely related to the number of samples. The larger the number of samples, the higher the accuracy of the prediction results. In addition, due to the effect of thermo-optic effect and thermal expansion effect, the change of ambient temperature will cause small changes in the size parameters of the microcavity, and the air flow may affect the coupling strength. In this example, multiple samplings (50 groups) are performed on the multi-resonant mode spectrum under the same large and small force within a short period of time to reduce the influence of temperature fluctuations and air flow on the prediction accuracy. Figure 6 shows the optical barcodes corresponding to different forces. Obviously, the barcode to be tested is the closest to the barcode corresponding to the pull force of 1.8N in the standard library, and the predicted value is around 1.8N.

Fig. 7 is a comparison chart of the predicted value and the real value obtained by predicting 50 test samples by the regression model after training. Among the 50 test samples, each 5 corresponds to the same force, 10 groups in total. Plot the true value of the test sample in the same coordinate system as the regression model's predicted value for its output. For the regression prediction model, the index to measure the fitting effect of the regression model is called the determination coefficient R ² , and the calculation formula is as follows:

为输出力的回归值，y为输出力的平均值，y_i为实验值。一般认为判定系数R²大于0.75时，表示模型拟合度较好，可解释程度较高。本实施例中的R²达到了0.98899，说明回归模型拟合效果好，预测结果精度高。In the formula, SST is the total sum of squares, SSR is the regression sum of squares,

is the regression value of the output force, y is the average value of the output force, and y _i is the experimental value. It is generally believed that when the determination coefficient ^R2 is greater than 0.75, it indicates that the model fit is better and the degree of interpretation is higher. The R ² in this example reached 0.98899, indicating that the regression model has a good fitting effect and the prediction result has a high precision.

Through the linear fitting of the 10 sets of small force prediction values output by the regression model in Figure 7 and the frequency shift, the linear fitting curve is obtained as:

ν=2.2861F-0.9449

In the fitting curve, F is the output force (N), and ν is the frequency shift (GHz). Then the sensitivity of the device is 2.2861 GHz/N. The minimum resolvable frequency shift of the regression model is 1.2×10 ^-5 GHz, so the calculated resolution of the system is 5.2μN.

Fig. 8 is a comparison diagram of the fitting curve of the predicted value of the tiny force and the fitting curve of the actual value. Each data point in Figure 8 is the experimental data corresponding to the predicted value, the solid line is the fitting line of the experimental data, and the dotted line is the curve corresponding to the real value. It can be seen that the fitting curve of the predicted value of the tiny force is very close to the fitting curve of the real value. It shows that the model can predict accurate data stably.

Figure 9 is the error histogram of the regression model, showing the error between the predicted output and the target output. The abscissa represents the error between the target output and the predicted output; the ordinate represents the number of outputs corresponding to the error size, and the zero error bar indicates that there is no deviation between the target output and the predicted output. It can be seen from Figure 9 that most of the error values are distributed around zero, and very few errors deviate from the zero point. The reasons for the error deviation from the zero point include the change of laboratory temperature, the small vibration of the platform caused by the experiment, and the air flow. However, the error between the overall data prediction output and the target output is extremely small, the prediction output accuracy is high, the model fitting effect is good, and the prediction result is accurate.

Figure 10 shows the regression prediction results of optical barcodes corresponding to 10 pull forces of different sizes. The trained neural network can accurately predict the absolute force corresponding to the barcode to be tested.

In summary, the present invention proposes a whispering gallery mode optical microcavity micro force sensor based on machine learning recognition calculation, which solves the problem that the single mode cannot be tracked when it moves out of the scanning range, and improves the applicability of the micro force sensor . The way to achieve it is to use machine learning algorithm to analyze the overall feature quantity within the pattern scanning range, and realize a large dynamic measurement range, real-time absolute micro force sensing mechanism, which saves manpower and material resources compared with manual calculation, and at the same time It also improves the speed and accuracy of measurement. At the same time, the micron-scale device size facilitates packaging and integration.

The basic principles, main features and advantages of the present invention have been shown and described above. Those skilled in the industry should understand that the present invention is not limited by the above-mentioned embodiments. What are described in the above-mentioned embodiments and the description only illustrate the principle of the present invention. Without departing from the spirit and scope of the present invention, the present invention will also have Variations and improvements are possible, which fall within the scope of the claimed invention. The protection scope of the present invention is defined by the appended claims and their equivalents.

Claims (9)

1. The micro-force sensing device based on optical microcavity utilizing machine learning is characterized in that: it comprises a microcavity microforce sensing and demodulation system, and the microcavity microforce sensing and demodulation system consists of a laser, a polarization controller, a coupling Composed of waveguides, optical microcavities, photodetectors, high-speed oscilloscopes, and data processing and machine learning modules; Among them, the laser generates continuous tunable laser light, the continuous tunable laser light is used as the input light, the polarization controller controls the polarization state of the input light, and the coupling waveguide couples the light into the optical microcavity; the optical microcavity is used as a high-sensitivity force sensor device to form a A series of high-Q multi-resonant modes; the optical microcavity is equipped with a piezoelectric fine adjustment module and a clamping coarse adjustment module. The piezoelectric fine adjustment module is used to load the micro force to be measured, and the clamping coarse adjustment module is used to fix the optical microcavity. and coarsely adjust the position of the microcavity; the photodetector is used to receive the optical signal from the optical microcavity coupling system, and convert it into an electrical signal and input it to a high-speed oscilloscope, which is used to monitor the resonant mode spectrum of the microcavity; data processing and machine learning The module is used to collect and process pattern spectrum data, and perform machine learning and analysis processing on pattern spectrum data. 2. The micro force sensing device based on optical microcavity utilizing machine learning according to claim 1, characterized in that: the laser is a continuous wavelength tunable laser, which is used in conjunction with an external signal generator for Injects a frequency-swept narrowband laser signal into the system. 3. The micro force sensing device based on optical microcavity using machine learning according to claim 1, characterized in that: the coupling waveguide and the optical microcavity form an Add-pass coupling structure or an Add-drop coupling structure. 4. according to the microforce sensing device based on the optical microcavity of utilizing machine learning described in claim 1, it is characterized in that: described optical microcavity adopts the optical microcavity of whispering gallery mode, solid or hollow, and profile is spherical, Neck-shaped, cylindrical or ellipsoidal. 5. The micro-force sensing device based on optical microcavity utilizing machine learning according to claim 1, characterized in that: the piezoelectric fine-tuning module adopts a piezoelectric ceramic stacking method, and by changing the voltage of the PZT, the The microcavity exerts an adjustable axial force; under the action of the driving force, the frequency shift of the multi-resonant mode spectrum of the microcavity occurs. 6. The micro force sensing device based on optical microcavity utilizing machine learning according to claim 5, characterized in that: said data processing and machine learning module process the collected multi-resonant mode spectral data into normalized A unified two-dimensional optical barcode, the optical barcode contains the spectral interval, line width, resonance wavelength and coupling depth information of multi-resonant modes. 7. The micro force sensing device based on optical microcavity utilizing machine learning according to claim 6, characterized in that: the optical barcode is composed of many rectangular areas, each rectangular area is a "bit" of the barcode ”, the width of each rectangular area represents the line width of the mode, and the color reflects its coupling depth, thus generating an optical barcode uniquely corresponding to the magnitude of the pulling force. 8. according to the micro-force sensing device based on optical microcavity that utilizes machine learning described in claim 7, it is characterized in that: described data processing and machine learning module collect the multi-resonant peak mode spectral data under different conditions, and The standard two-dimensional optical barcodes corresponding to different sizes of forces are obtained by processing; a large number of standard optical barcodes are used as input, and the micro-force size and stretching length under corresponding conditions are used as output. After training with neural network algorithms, the input data can be predicted. Large and small convolutional neural network regression model; apply the force to be measured to the optical microcavity coupling system, obtain the optical barcode to be tested after spectral collection and data processing, and input the two-dimensional optical barcode image to the trained neural network model for regression Analyze, obtain the regression value of the force to be measured, and obtain the magnitude of the absolute force. 9. According to the micro-force sensing device based on optical microcavity utilizing machine learning described in any one of claims 1-8, it is characterized in that the operating principle of the micro-force sensing device is: First, the laser generates a continuous frequency-sweeping laser signal, which is coupled into the optical microcavity after the polarization state is adjusted by the polarization controller; due to the ultra-high quality factor of the optical microcavity, the coupled output optical signal forms a high-Q multi-resonance mode spectrum; A large number of spectrograms under the condition of measuring force are input into the machine learning model for training and testing, and a neural network model that can identify the spectral similarity of resonance modes is established; the spectrum of multi-resonance modes under the action of the force to be measured is collected and imported into the neural network after successful training. Network model, so as to output the magnitude of the force to be measured.

Images