High-precision fast-response micro-nano optical fiber temperature and humidity sensor combined with machine learning auxiliary demodulation
Through parallel sparse spectral sampling and machine learning-assisted demodulation methods, the problem of complex and cost of demodulation of fiber temperature and humidity sensors in harsh environments is solved, and a high-precision and fast response fiber temperature and humidity sensor is realized, which is suitable for real-time monitoring of complex environments.
Patent Information
- Application Number
- CN202510433180.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-08
- Publication Date
- 2025-08-01
AI Technical Summary
The existing fiber optic temperature and humidity sensors have complex and high cost in harsh environments such as high temperature and high voltage and strong electromagnetic interference, making it difficult to achieve high-precision and fast response.
Using parallel sparse spectral sampling and machine learning methods, 1×N photodetector array and artificial intelligence data processing unit are used to directly predict temperature and humidity through trained artificial neural networks, replacing traditional full-spectral scanning.
It realizes MHz-level response speed and high-precision demodulation, reducing system complexity and cost, and improving the practicality of the sensor.
Smart Images
Figure CN120403728A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of fiber optic sensing, and particularly relates to a high-precision and fast-response micro-nano fiber temperature and humidity sensor combined with machine learning-assisted demodulation. The sensor realizes fast and accurate discrimination detection of temperature and humidity through parallel sparse spectral sampling and artificial intelligence algorithms, and is applicable to real-time monitoring scenarios in complex environments (such as high temperature and high pressure, strong electromagnetic interference, etc.). Background Art
[0002] The temperature and humidity changes of gases are important monitoring parameters for the atmospheric environment, industrial production processes, and human health status. Although traditional electronic temperature and humidity sensors are widely used, they cannot work properly in harsh environments such as high temperature and high pressure, high corrosion, and strong electromagnetic interference. Fiber optic sensors have the advantages of anti-electromagnetic interference and anti-corrosion, and have become an alternative solution. Among them, micro-nano fibers can significantly increase the interaction area between light and the medium to be measured due to the evanescent field enhancement effect, thereby improving the sensing sensitivity. Research shows that micro-nano fiber gas sensors can introduce a detectable optical path difference within a working distance of a few millimeters, making them an ideal platform for high-precision temperature and humidity sensing. In 2022, D. Gao et al. prepared a micro-nano fiber humidity sensor that can compensate for temperature drift. The humidity sensitivity of the sensor is -47 pm / %RH, and the waist region length of the sensor is 2-3 mm (D. Gan, et al., A Relative Humidity Sensor Based on Non-Adiabatic Tapered Optical Fiber for Remote Measurement in Power Cable Tunnel, IEEE Transactions on Instrument and Measurement, 2022, 71). However, for this kind of fiber optic temperature and humidity sensor based on wavelength modulation, traditional demodulation methods need to track two characteristic wavelengths and use a cross-sensitivity matrix for discrimination and analysis. To ensure the accuracy of characteristic wavelength positioning, an expensive spectrometer is required for high-resolution spectral scanning in the experiment, and complex spectral processing algorithms are used in combination. When the sensitivity of the sensor is high or the distance between the two characteristic wavelengths is large, the time required for a single spectral scan may be several seconds, which greatly limits the application of fiber optic temperature and humidity sensors in fast-response scenarios (such as human breath gas monitoring).
[0003] To improve the response ability of wavelength modulation type fiber optic sensors, high-speed spectrometers have emerged on the market, but they are expensive. In 2021, Zhang et al. used a wavelength division multiplexer to split the broadband spectrum into coarse spectra to achieve fast demodulation of interferometric sensor signals (Zhang.P, et al., A High-Speed Demodulation Technology of Fiber Optic Extrinsic Fabry-Perot Interferometric Sensor Based on Coarse Spectrum. Sensors, 2021, 21). However, while this method improves the demodulation speed, it increases the complexity of the system.
[0004] Therefore, how to reduce costs and improve demodulation efficiency while ensuring accuracy has become the key challenge in promoting the practical application of fiber optic temperature and humidity sensors. In addition, existing methods rely heavily on hardware and it is difficult to balance high integration and low cost. Summary of the Invention
[0005] To solve the problems of the existing technology, the purpose of the present invention is to provide a high-precision and fast-response micro-nano fiber optic temperature and humidity sensor with machine learning-assisted demodulation. By parallel sparse spectral sampling and machine learning methods, potential features in the sparse spectral sampling points are extracted to replace traditional full-spectrum scanning, achieving an MHz-level response speed and high-precision demodulation while reducing the system complexity.
[0006] To achieve the above object of the invention, the present invention adopts the following technical solutions:
[0007] A high-precision and fast-response micro-nano fiber optic temperature and humidity sensor with machine learning-assisted demodulation, characterized in that the sensing system includes a broadband light source, an input optical fiber, a micro-nano fiber optic temperature and humidity sensor, an output optical fiber, an N-channel optical narrowband filter, a 1×N photodetector array, and an artificial intelligence data processing unit.
[0008] The laser with continuous wavelength emitted by the broadband light source is transmitted to the micro-nano fiber optic temperature and humidity sensor through the input optical fiber; the light output by the micro-nano fiber optic temperature and humidity sensor is transmitted to the N-channel optical narrowband filter through the output optical fiber. The 1×N photodetector array is connected correspondingly to the output of the N-channel optical narrowband filter. The N light intensity values output by the photodetector array are input in parallel to the artificial intelligence data processing unit. The artificial intelligence data processing unit contains a trained artificial neural network, which directly predicts the sensed temperature and humidity using the N light intensity values.
[0009] The artificial neural network in the artificial intelligence data processing unit is trained using the following technical solutions:
[0010] Step 1: Connect the interference spectrum output by the broadband light source through the micro-nano fiber temperature and humidity sensor to a spectrometer, and collect several groups of interference spectra under different temperature and humidity conditions;
[0011] Step 2: Perform generative adversarial network data augmentation on the collected interference spectra to obtain multiple groups of pseudo-spectra with similar but different data distributions;
[0012] Step 3: Use the sparse spectrum sampling technique to scan the original spectrum and the generated pseudo-spectra of the micro-nano fiber temperature and humidity sensor at intervals, and obtain m groups of sparse spectra containing N sampling points (consistent with the number of detectors) under different parameters as data samples. Then divide the sample data set into a training data set and a test data set according to an appropriate data ratio;
[0013] Step 4: Use the data samples obtained in Step 3 to train the artificial neural network model H; the neural network includes an input layer whose nodes correspond to N spectral sampling points; several hidden layers, and an output layer whose nodes correspond to temperature and humidity; the artificial neural network uses Bayesian regularization to optimize the parameters, and uses the comprehensive objective of the minimum mean square error (MSE) and network weights as the network training objective.
[0014] The N light intensity values output by the 1×N photodetector array are input into the trained network model H to obtain the predicted values of temperature and humidity
[0015] Compared with the prior art, the present invention has the following obvious outstanding substantive features and remarkable advantages:
[0016] 1. The present invention uses the N light intensity values output by the 1×N photodetector array for parallel sampling, and uses a machine learning-assisted wavelength demodulation method to directly obtain the predicted values of temperature and humidity; the acquisition time of a single spectral data required by the wavelength demodulation method of the present invention is only the acquisition time of a spectral sampling point data in the traditional method, and can realize fast-response (MHz-level) fiber temperature and humidity sensing.
[0017] 2. The present invention uses the N light intensity values output by the 1×N photodetector array for parallel sampling to replace the full spectrum collected by the spectrometer for wavelength demodulation, and can realize a low-cost and high-integration fiber temperature and humidity sensing system without reducing the demodulation accuracy. BRIEF DESCRIPTION OF THE DRAWINGS
[0018] Figure 1 is a hardware schematic diagram of an embodiment of a high-precision and fast-response micro-nano fiber temperature and humidity sensor combined with machine learning-assisted demodulation of the present invention.
[0019] Figure 2 is a data processing flowchart of a high-precision and fast-response micro-nano fiber temperature and humidity sensor combined with machine learning-assisted demodulation of the present invention.
[0020] Figure 3 are the original spectrum and the parallel sparse sampling spectrum of the fiber optic sensor according to the preferred embodiment of the present invention.
[0021] Figure 4 are the result diagrams of the model training set and the test set according to the preferred embodiment of the present invention. Detailed implementation manners
[0022] The following describes the detailed implementation manners of the present invention with reference to the accompanying drawings, so that those skilled in the art can better understand the present invention. It should be particularly noted that in the following description, when the detailed description of known functions and designs may dilute the main content of the present invention, these descriptions will be omitted here.
[0023] Embodiment:
[0024] Specifically, it relates to a high-precision and fast-response micro-nano fiber temperature and humidity sensor combined with machine learning-assisted demodulation, and its hardware system is constructed as Figure 1 shown, including:
[0025] A broadband light source 1. In this embodiment, a supercontinuum light source (such as an erbium-doped fiber light source) is selected, with a wavelength range covering the sensitive band of the micro-nano fiber sensor and a stable output power.
[0026] An input optical fiber 2 and an output optical fiber 4, using single-mode optical fibers to ensure low-loss transmission of optical signals.
[0027] A micro-nano fiber temperature and humidity sensor 3. The micro-nano fiber is prepared by the fused biconical taper process, with a waist region length of 2 - 3 mm and a diameter of about 2 - 3 μm to enhance the evanescent field effect.
[0028] An N-channel optical narrowband filter 5, using a multi-channel filtering module based on an arrayed waveguide grating (AWG) or a Fabry-Perot (FP) filter. The number of channels N is set according to the spectral feature distribution (such as N = 5), and the bandwidth of each channel is about 0.5 nm, covering the characteristic wavelength offset range of the sensor to ensure that key feature points are captured. In this embodiment, the light intensity values at specific wavelength points are collected, such as 1520 nm, 1550 nm, 1580 nm, etc. The high-resolution sampling spectrum is as Figure 3 (a) shown, and the parallel sampled data points are as Figure 3 (b) shown, and the data acquisition time is shortened to the microsecond level.
[0029] A 1×N photodetector array 6, using high-speed InGaAs photodiodes (response time ≤ 1 ns), with each detector corresponding to a narrowband channel, and the output light intensity signal is amplified and then transmitted to the data processing unit.
[0030] The artificial intelligence data processing unit 7 integrates an FPGA or an embedded GPU module, runs a pre-trained artificial neural network model, and outputs temperature and humidity prediction values in real time.
[0031] The workflow is as follows:
[0032] The laser with a continuous wavelength emitted by the broadband light source (1) is transmitted to the micro-nano fiber temperature and humidity sensor (3) through the input optical fiber (2). The environmental temperature and humidity changes cause the wavelength shift of the interference spectrum of the sensor. The modulated optical signal is output to the N-channel optical narrowband filter (5) through the output optical fiber (4) and decomposed into N narrowband spectra. The output ends of the narrowband filters (5) of each channel are correspondingly connected to the input ends of the 1×N photodetector array (6). The 1×N photodetector array (6) synchronously collects the light intensity values of each channel, forms N light intensity values and inputs them into the artificial intelligence data processing unit (7). The artificial intelligence unit calculates the temperature and humidity values in real time based on the N light intensity values through the neural network model and outputs the predicted values of temperature and humidity.
[0033] Figure 2 It is the data processing flow chart of the high-precision and fast-response micro-nano fiber temperature and humidity sensor combined with machine learning-assisted demodulation in the present invention. As shown in the figure, it includes the following steps:
[0034] Step 1: Interference spectrum acquisition
[0035] Place the sensor in a temperature and humidity controllable cavity (such as a thermostatic and humidistatic chamber), and use a high-resolution spectrometer (resolution ≤ 0.01 nm) to collect the interference spectra of the sensor under different temperature and humidity combinations, and collect about 100 groups of data in total
[0036] Step 2: Generative adversarial network (GAN) data augmentation
[0037] Adopt a generative adversarial network (GAN). The input of the generator is random noise, and the input of the discriminator is the real / generated spectrum. Generate pseudo-spectra with the same distribution as the real spectrum but different wavelength offsets, expand the data set to 500 groups, and enhance the generalization ability of the model.
[0038] Step 3: Sparse spectrum sampling
[0039] According to the characteristic wavelength distribution, uniformly select N sampling points (consistent with the number of detectors) of the sparse spectrum in the original spectrum as data samples. Each group of data contains the light intensity values of N sampling points and the corresponding temperature and humidity labels, forming an m×N matrix.
[0040] Step 4: Artificial neural network training
[0041] 4.1 Model structure design:
[0042] Input layer: N nodes, corresponding to the light intensity values of N sampling points.
[0043] Hidden layer: There are several hidden layers. In this embodiment, a 3-layer fully connected network is used, with the number of nodes being 128, 64, and 32 respectively. The ReLU activation function is adopted.
[0044] Output layer: There are 2 nodes, corresponding to temperature and humidity respectively, outputting temperature (°C) and humidity (%RH). The activation function is linear.
[0045] Regularization: Bayesian regularization is adopted, and a weight L2 norm penalty term is introduced into the loss function to prevent overfitting.
[0046] 4.2 Training process
[0047] The sample data set is divided into a training data set, a validation set, and a test data set according to an appropriate data ratio. For example, it is divided into a training set (400 groups), a validation set (50 groups), and a test set (50 groups) in the ratio of 8:1:1;
[0048] Minimize the mean square error (MSE);
[0049] Use the Adam optimizer, with an initial learning rate of 0.001, a batch size of 64, and 500 iterations;
[0050] The experimental results are as Figure 4 shown, Figure 4 (a) and (c) are respectively the results of temperature and humidity analysis of the training set data, Figure 4 (b) and Figure 4 (d) are respectively the results of temperature and humidity analysis of the test set data. The experimental data shows that the average relative error (MRE) of temperature is 0.258%, and the average relative error (MRE) of humidity is 0.659%.
[0051] In this embodiment, the hardware complexity is reduced by sparse sampling, and then the resolution loss is compensated by a machine learning model. Finally, the detection accuracy equivalent to that of a traditional high-resolution spectrometer is achieved on the premise of low cost and high speed. While ensuring high accuracy, the response speed and practicability of the fiber optic temperature and humidity sensor are improved, providing an efficient solution for complex environment monitoring.
[0052] The above has described the embodiments of the present invention in conjunction with the accompanying drawings. However, the present invention is not limited to the above embodiments, and various changes can be made according to the purpose of the invention of the present invention. Any changes, modifications, substitutions, combinations, or simplifications made based on the spirit and principle of the technical solution of the present invention shall be equivalent replacement methods, as long as they conform to the invention purpose of the present invention and do not deviate from the technical principle and inventive concept of the present invention, they all belong to the protection scope of the present invention.
Claims
1. A high-precision and fast-response micro-nano fiber temperature and humidity sensor combined with machine learning-assisted demodulation, characterized in that, Comprising: A broadband light source (1) for generating a laser with a continuous wavelength; An input optical fiber (2) connected to the broadband light source (1) for transmitting the laser to the micro-nano fiber temperature and humidity sensor (3); A micro-nano fiber temperature and humidity sensor (3) for responding to environmental temperature and humidity changes and modulating and outputting an optical signal; An output optical fiber (4) connected to the micro-nano fiber temperature and humidity sensor (3) for transmitting the modulated optical signal; An N-channel optical narrowband filter (5) connected to the output optical fiber (4) for decomposing the optical signal into N narrowband spectral channels; A 1×N photodetector array (6) correspondingly connected to the output end of the N-channel optical narrowband filter (5) for synchronously collecting the light intensity values of each channel; An artificial intelligence data processing unit (7) connected to the photodetector array (6), including a pre-trained artificial neural network model for real-time predicting temperature and humidity values according to the N light intensity values.
2. The high-precision and fast-response micro-nano optical fiber temperature and humidity sensor combined with machine learning-assisted demodulation according to claim 1, characterized in that The micro-nano fiber temperature and humidity sensor (3) is a non-adiabatic tapered micro-nano fiber, and its waist diameter satisfies that there are several free spectral ranges near the dispersion turning point within the wavelength range of the broadband light source (1), and the dispersion turning point satisfies where Δn eff12 is the effective refractive index difference between the modes HE 11 and HE 12 in the micro-nano fiber.
3. The high-precision and fast-response micro-nano fiber temperature and humidity sensor combined with machine learning-assisted demodulation according to claim 1, wherein The training method of the artificial neural network includes the following steps: Step 1): Collecting a plurality of groups of original interference spectra under different temperature and humidity conditions through a spectrometer; Step 2): Using a generative adversarial network (GAN) to perform data augmentation on the original spectrum to generate multiple groups of pseudo-spectra with similar but different data distributions to expand the data set; Step 3): Performing sparse spectral sampling on the original interference spectrum and the pseudo-spectra to obtain N sampling point light intensity values at m different parameters consistent with the number of detectors; Step 4): Training the artificial neural network model based on the sampling point light intensity values, and the optimization objective is to minimize the mean square error (MSE) of the predicted temperature and humidity.
4. The high-precision and fast-response micro-nano fiber temperature and humidity sensor combined with machine learning-assisted demodulation according to claim 3, characterized in that, The structure of the artificial neural network includes: An input layer with N nodes corresponding to the N sampling point light intensity values; At least two hidden layers with the activation function ReLU; An output layer containing two nodes respectively outputting temperature and humidity values; During the training process, Bayesian regularization is used to optimize the network weights.
5. The high-precision and fast-response micro-nano optical fiber temperature and humidity sensor combined with machine learning-assisted demodulation according to claim 1, characterized in that, The micro-nano fiber temperature and humidity sensor (3) is prepared by a fused biconical taper process, with a waist region length of 2-3 mm and a diameter of 2-3 μm.
6. The high-precision and fast-response micro-nano fiber temperature and humidity sensor combined with machine learning-assisted demodulation according to claim 1, wherein, The artificial intelligence data processing unit (7) integrates an FPGA or an embedded GPU module, supports real-time data processing, and the system response time ≤1 μs.
7. The high-precision and fast-response micro-nano fiber temperature and humidity sensor combined with machine learning-assisted demodulation according to claim 1, characterized in that The N-channel optical narrowband filter (5) is an arrayed waveguide grating or a Fabry-Perot filter, with a channel interval of 10-50 nm and a single-channel bandwidth ≤1 nm; the 1×N photodetector array (6) uses a high-speed InGaAs photodiode, with a response time ≤1 ns, and the light intensity signal is amplified and then input into the artificial intelligence data processing unit.
8. The high-precision and fast-response micro-nano fiber temperature and humidity sensor combined with machine learning-assisted demodulation according to claim 1, characterized in that The interval Δ of the sparse spectral sampling can reach 20-40 nm, covering the sensor characteristic wavelength offset range of ±15 nm.
9. The high-precision and fast-response micro-nano fiber temperature and humidity sensor combined with machine learning-assisted demodulation according to claim 1, wherein A low-pass filter is arranged at the front end of the photodetector array (6), with a cut-off frequency ≤1 MHz, for suppressing high-frequency noise.