An intelligent sensor-based asphalt regeneration performance evaluation method

Abstract

The present application relates to the technical field of intelligent sensor, in particular to a method for evaluating asphalt regeneration performance based on intelligent sensor. The probe integrating near-infrared light source, miniature spectrometer, high-density micro-thermal resistance array and edge computing chip is placed in the interior of the measured regeneration asphalt, the light source is controlled to emit light and the near-infrared diffuse reflectance spectrum data of the asphalt is collected, and the micro-area temperature change sequence in the light source alternating on-off cycle is synchronously collected; the absorption peak position offset of the characteristic functional group of the regenerant in the spectrum is extracted, and the heat conduction time constant in the temperature sequence is calculated; the offset and the time constant are input into the multi-dimensional feature fusion network in the edge computing chip, and the diffusion uniformity index and compatibility grade of the regenerant in the aged asphalt are output. The present application synchronously obtains the spectrum and heat conduction characteristics in situ and carries out fusion analysis, eliminates the damage of offline sampling to the original structure of the sample, and realizes real-time evaluation of the micro-diffusion dynamic process of the regenerant.

Description

Technical Field

[0001] This invention relates to the field of intelligent sensor technology, and specifically to a method for evaluating the performance of asphalt recycling based on intelligent sensors. Background Technology

[0002] In asphalt recycling technology, evaluating the compatibility of the recycling agent with aged asphalt is a crucial step in determining the recycling agent dosage and construction process. Currently, the commonly used technical approach in the industry is the sampling laboratory testing method. Specifically, samples of asphalt mixtures with added recycling agent are obtained at the construction site or mixing plant. These samples are then sealed and transferred to a testing laboratory located far from the construction site. Macroscopic physical and mechanical properties of the samples are tested using a penetration tester, softening point meter, ductility tester, or dynamic shear rheometer. By comparing the macroscopic mechanical data obtained from the tests with thresholds in standard specifications, testing personnel indirectly infer the dispersion state of the recycling agent in the aged asphalt and the degree of compatibility between the two. The entire testing process relies on the acquisition of final-state data of the macroscopic mechanical properties.

[0003] The aforementioned conventional technical solutions require stripping asphalt samples from the construction environment and transferring them to a laboratory environment during testing. This offline sampling and testing method cannot preserve the true structure of the asphalt in its original construction state. Since macroscopic physical and mechanical index tests obtain the final static data after the recycling agent and aged asphalt have been mixed, the test results mask the dynamic process of diffusion and dissolution of the recycling agent at the microscopic level within the aged asphalt. When the recycling agent only undergoes physical encapsulation without microscopic dissolution and penetration, the macroscopic mechanical indexes may present a false appearance of compliance, making it impossible for existing technologies to reflect the true microscopic compatibility state within the recycled asphalt in real time. Summary of the Invention

[0004] The purpose of this invention is to provide a method for evaluating the performance of asphalt recycling based on intelligent sensors, which can solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A method for evaluating the performance of recycled asphalt based on intelligent sensors is disclosed. The method is applied to an intelligent sensor probe comprising a near-infrared light source, a miniature spectrometer, a high-density micro-thermometer array, and an edge computing chip. The method includes: the edge computing chip controlling the near-infrared light source to emit a detection beam at a preset frequency; the miniature spectrometer acquiring near-infrared diffuse reflectance spectral data of the recycled asphalt to be tested; the high-density micro-thermometer array synchronously acquiring the temperature change sequence of a micro-region of the recycled asphalt to be tested during the alternating on and off cycles of the near-infrared light source; the edge computing chip extracting the absorption peak position offset of the corresponding characteristic functional groups of the recycler in the near-infrared diffuse reflectance spectral data and calculating the thermal conductivity time constant in the temperature change sequence; the edge computing chip inputting the absorption peak position offset and the thermal conductivity time constant into a multi-dimensional feature fusion network pre-installed within the edge computing chip, and outputting the diffusion uniformity index and compatibility level of the recycler in aged asphalt.

[0006] Preferably, the edge computing chip controls the near-infrared light source to emit a detection beam at a preset frequency, including: the edge computing chip generates a pulse modulation signal with a first timing sequence, and drives the near-infrared light source to alternately output a first center wavelength beam and a second center wavelength beam according to the pulse modulation signal with the first timing sequence; the micro spectrometer collects near-infrared diffuse reflectance spectral data of the recycled asphalt to be tested, including: the diffraction grating built into the micro spectrometer alternately switches according to the pulse modulation signal with the first timing sequence; the micro spectrometer collects first band spectral data during the illumination of the first center wavelength beam, and collects second band spectral data during the illumination of the second center wavelength beam; the edge computing chip stitches the first band spectral data and the second band spectral data into the near-infrared diffuse reflectance spectral data.

[0007] Preferably, the high-density micro-thermistor array synchronously acquires the temperature change sequence of the micro-area of ​​the recycled asphalt under test during the alternating on and off cycle of the near-infrared light source, including: each micro-thermistor in the high-density micro-thermistor array is connected to a corresponding Wheatstone bridge; the edge computing chip records the first output voltage value of the Wheatstone bridge when the near-infrared light source is turned on, and records the second output voltage value of the Wheatstone bridge when the near-infrared light source is turned off; the edge computing chip generates the temperature change sequence of the micro-area where each micro-thermistor is located based on the difference sequence between the first output voltage value and the second output voltage value, wherein the high-density micro-thermistor array is distributed in a ring around the periphery of the near-infrared light source.

[0008] Preferably, the edge computing chip extracts the absorption peak position offset of the corresponding regenerant characteristic functional group in the near-infrared diffuse reflectance spectral data, including: the edge computing chip performs polynomial fitting on the near-infrared diffuse reflectance spectral data to obtain a baseline curve, subtracts the baseline curve from the near-infrared diffuse reflectance spectral data to obtain corrected spectral data, the edge computing chip determines a preset wavenumber interval corresponding to the regenerant characteristic functional group in the corrected spectral data, finds the wavenumber corresponding to the minimum point in the preset wavenumber interval as the measured absorption peak position, and subtracts the measured absorption peak position from the pre-stored standard absorption peak position to obtain the absorption peak position offset.

[0009] Preferably, the edge computing chip calculates the heat conduction time constant in the temperature change sequence, including: the edge computing chip extracts cooling segment data from the temperature change sequence starting from the moment the near-infrared light source is turned off; discretizes the cooling segment data to obtain a discrete temperature point set; the edge computing chip performs least-squares fitting on the discrete temperature point set using a preset exponential decay function; the reciprocal of the coefficient of the fitted exponential term is determined as the heat conduction time constant; and the edge computing chip calculates the average value of the heat conduction time constant corresponding to each micro-thermistor in the high-density micro-thermistor array as the final heat conduction time constant.

[0010] Preferably, the edge computing chip inputs the absorption peak position offset and the thermal conduction time constant into a multi-dimensional feature fusion network pre-installed within the edge computing chip, including: the edge computing chip performing maximum and minimum normalization processing on the absorption peak position offset and the thermal conduction time constant respectively; concatenating the normalized absorption peak position offset and the normalized thermal conduction time constant into a two-dimensional input vector; inputting the two-dimensional input vector into a fully connected layer of the multi-dimensional feature fusion network; the fully connected layer performing matrix multiplication and bias addition operations on the two-dimensional input vector and then inputting it into an activation function layer; and the activation function layer outputting the diffusion uniformity index and the compatibility level.

[0011] Preferably, before the edge computing chip stitches the first band spectral data and the second band spectral data into the near-infrared diffuse reflectance spectral data, the method further includes: the edge computing chip obtaining the current temperature value of the ambient temperature sensor inside the micro spectrometer, looking up the current temperature value in a pre-stored temperature drift compensation table to obtain a first compensation coefficient corresponding to the first band spectral data and a second compensation coefficient corresponding to the second band spectral data, and the edge computing chip dividing the light intensity value of each wavelength in the first band spectral data by the first compensation coefficient, and dividing the light intensity value of each wavelength in the second band spectral data by the second compensation coefficient.

[0012] Preferably, each micro-thermistor in the high-density micro-thermistor array is connected to a corresponding Wheatstone bridge, including: the Wheatstone bridge includes the micro-thermistor, a fixed resistor, a first matching resistor, and a second matching resistor, wherein the first matching resistor and the second matching resistor are reference thermistors with the same temperature coefficient as the micro-thermistor, the reference thermistor is encapsulated in a heat-insulating cavity inside the smart sensor probe to isolate the recycled asphalt to be tested, the edge computing chip applies a constant current source excitation to the Wheatstone bridge, and the first output voltage value and the second output voltage value are the voltage difference between the two ends of the diagonal of the Wheatstone bridge.

[0013] Preferably, the edge computing chip performs polynomial fitting on the near-infrared diffuse reflectance spectral data to obtain a baseline curve, including: the edge computing chip divides the near-infrared diffuse reflectance spectral data into multiple consecutive wavenumber windows; for each wavenumber window, the edge computing chip sequentially uses a quadratic polynomial, a cubic polynomial, and a quartic polynomial to fit the spectral data within the current wavenumber window; calculates the sum of squared residuals after fitting each polynomial; selects the polynomial with the smallest sum of squared residuals as the target polynomial for the current wavenumber window; and generates a continuous baseline curve based on the target polynomial corresponding to each wavenumber window.

[0014] Preferably, the activation function layer outputs the diffusion uniformity index and the compatibility level, including: the activation function layer outputs a first value and a second value; the edge computing chip inputs the first value into a pre-stored first piecewise linear mapping function, determines the output value of the first piecewise linear mapping function as the diffusion uniformity index; the edge computing chip inputs the second value into a pre-stored second piecewise linear mapping function, compares the output value of the second piecewise linear mapping function with a plurality of pre-stored compatibility level thresholds, and outputs the label corresponding to the compatibility level threshold that matches the comparison as the compatibility level.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention utilizes an intelligent sensor probe integrating a near-infrared light source, a miniature spectrometer, a high-density micro-thermometer array, and an edge computing chip. It simultaneously acquires near-infrared diffuse reflectance spectral data and temperature change sequences within the recycled asphalt under test. The absorption peak position shifts and thermal conduction time constants of characteristic functional groups are extracted and input into a multi-dimensional feature fusion network pre-installed within the edge computing chip. This enables in-situ, real-time output of the diffusion uniformity and compatibility level of the recycling agent in aged asphalt. This scheme performs feature-level fusion of microscopic spectral characteristic absorption frequency shifts and micro-area thermal conduction delay data, directly analyzing the microscopic diffusion and dissolution state of the recycling agent in situ during construction. This eliminates the damage to the original asphalt structure caused by offline sampling and solves the problem that existing technologies relying on macroscopic mechanical final-state data cannot reflect the dynamic process of microscopic compatibility in real time.

[0016] 2. In the spectral acquisition stage, this invention uses a first-sequence pulse modulation signal to drive a near-infrared light source to alternately output beams of different center wavelengths and synchronously switch diffraction gratings. In the temperature acquisition stage, a Wheatstone bridge with a ring distribution and a reference thermistor with the same temperature coefficient is used to acquire the output voltage difference sequence, eliminating common-mode interference from ambient temperature fluctuations on micro-area temperature acquisition. In the data processing stage, a baseline curve is generated by automatically matching the polynomial with the minimum sum of squared residuals within a divided wavenumber window to extract the absorption peak position offset. The least-squares fitting of the discrete temperature point set in the cooling section is performed using an exponential decay function to calculate the heat conduction time constant. The spectral intensity value is corrected by combining an ambient temperature lookup table compensation coefficient, filtering out environmental drift and background noise interference, and ensuring the accuracy of the input two-dimensional vector. Matrix operations and threshold comparisons are performed on the standardized features through a fully connected layer and a piecewise linear mapping function, converting microscopic physical quantities into quantified diffusion uniformity indicators and hierarchical labels, reducing the amount of raw sensor data transmitted. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the overall process of in-situ detection using the intelligent sensor probe of the present invention. Figure 2 This is a flowchart of the dual-wavelength pulse modulation spectral acquisition and correction process of the present invention; Figure 3 This is a flowchart of the high-density micro-thermometer array temperature acquisition process of the present invention; Figure 4 This is a flowchart of the spectral baseline correction and absorption peak shift extraction process of the present invention; Figure 5 This is a flowchart illustrating the calculation process of the heat conduction time constant of the present invention. Figure 6 This is a flowchart of the multi-dimensional feature fusion network inference output process of the present invention. Detailed Implementation

[0018] The asphalt recycling performance evaluation method of this invention is applied to an intelligent sensor probe, which integrates a near-infrared light source, a miniature spectrometer, a high-density micro-thermistor array, and an edge computing chip. The light-emitting window of the near-infrared light source faces the detection surface of the recycled asphalt to be tested, emitting a near-infrared detection beam. The receiving optical path of the miniature spectrometer forms a preset diffuse reflection angle with the light-emitting optical path of the near-infrared light source, collecting the near-infrared diffuse reflection light reflected from the surface of the recycled asphalt and converting it into spectral data. The detection surface of the high-density micro-thermistor array is attached to the detection surface of the recycled asphalt to collect temperature change data of micro-regions within the recycled asphalt. The edge computing chip is electrically connected to the near-infrared light source, the miniature spectrometer, and the high-density micro-thermistor array, respectively, and is used to output control commands, execute synchronous acquisition scheduling, and complete data processing calculations.

[0019] refer to Figure 1 In one embodiment, the edge computing chip generates a pulse drive signal at a preset frequency and transmits the pulse drive signal to the drive circuit of the near-infrared light source to control the near-infrared light source to emit a detection beam at the preset frequency. Specifically, the preset frequency can be configured according to the on-site testing environment of the recycled asphalt to be tested, and the duty cycle of the pulse drive signal can be set according to the energy requirements of the detection beam, so that the near-infrared light source completes one alternating on and off operation within one pulse cycle. The edge computing chip has a built-in hardware timer, the clock source of which is a high-precision crystal oscillator. The edge computing chip generates a synchronization trigger signal through the hardware timer and outputs it to the drive circuit of the near-infrared light source, the trigger terminal of the micro spectrometer, and the acquisition control terminal of the high-density micro thermistor array, respectively, to ensure that the beam emission of the near-infrared light source, the spectral acquisition of the micro spectrometer, and the temperature acquisition of the high-density micro thermistor array are completely synchronized in time, eliminating timing deviations between different acquisition links.

[0020] Furthermore, during the alternating on and off process of the near-infrared light source at a preset frequency, the edge computing chip synchronously schedules the micro-spectrometer and the high-density micro-thermometer array to perform data acquisition operations. The micro-spectrometer collects the near-infrared light reflected from the surface of the recycled asphalt under test through its built-in optical receiving components, converts the optical signal into an electrical signal, and generates the near-infrared diffuse reflectance spectrum data of the recycled asphalt under test. The detection surface of the high-density micro-thermometer array is directly attached to the micro-area surface of the recycled asphalt under test. During the cycle when the near-infrared light source is on, the micro-area of ​​the recycled asphalt under test absorbs the energy of the near-infrared light and generates temperature changes. During the cycle when the near-infrared light source is off, the micro-area of ​​the recycled asphalt under test generates temperature changes through heat conduction. During the complete cycle of the near-infrared light source alternating on and off, the high-density micro-thermometer array collects the temperature values ​​of the corresponding micro-area of ​​the recycled asphalt under test at a sampling frequency higher than the pulse drive signal frequency, generating a continuous temperature change sequence. The edge computing chip performs outlier removal on the acquired near-infrared diffuse reflectance spectral data and temperature change sequence. For sampling points in the spectral data whose light intensity values ​​exceed the preset reasonable range, the average light intensity values ​​of adjacent sampling points are used for replacement. For sampling points in the temperature change sequence whose temperature values ​​exceed the preset reasonable range, the median temperature values ​​of adjacent sampling points are used for replacement, thereby eliminating random noise and sudden interference during the acquisition process.

[0021] Furthermore, the edge computing chip performs feature extraction on the acquired near-infrared diffuse reflectance spectral data, extracting the position shift of the absorption peak corresponding to the characteristic functional groups of the regenerator. Specifically, the characteristic functional groups of the regenerator include, but are not limited to, saturated hydrocarbon CH bonds, aromatic CH bonds, hydroxyl OH bonds, and carbonyl C=O bonds. Different characteristic functional groups correspond to specific wavenumber ranges in the near-infrared spectrum. The shift in the absorption peak position corresponds to the change in hydrogen bonding and intermolecular forces between the regenerator molecules and aged asphalt molecules, which can characterize the dissolution state of the regenerator in aged asphalt. The edge computing chip performs baseline correction processing on the near-infrared diffuse reflectance spectral data to eliminate background noise and baseline drift interference during the spectral acquisition process. In the corrected spectral data, it locates the wavenumber range corresponding to the characteristic functional groups of the regenerator, identifies the measured position of the absorption peak within this wavenumber range, and calculates the difference between the measured position and the pre-stored standard absorption peak position to obtain the absorption peak position shift.

[0022] Specifically, Table 1 is a comparison table of characteristic functional groups of the regenerator and their corresponding preset wavenumber intervals. The preset wavenumber intervals and standard absorption peak positions corresponding to the characteristic functional groups of the regenerator listed in the table are pre-stored in the built-in storage unit of the edge computing chip for the location and offset calculation of the absorption peak positions. The standard absorption peak position is the absorption peak wavenumber of the corresponding characteristic functional group obtained by laboratory standard Fourier transform near-infrared spectrometer detection in the pure state of the regenerator.

[0023] Table 1. Comparison of Characteristic Functional Groups of Regenerants and Corresponding Preset Wavenumber Ranges

[0024] Furthermore, the edge computing chip performs parameter calculations on the acquired temperature change sequence to obtain the thermal conduction time constant within the temperature change sequence. Specifically, the thermal conduction time constant corresponds to the thermal diffusion rate of the micro-region of the recycled asphalt being tested. Different diffusion uniformities of the rejuvenator in the aged asphalt lead to differences in the thermal conductivity of the recycled asphalt, which is reflected in changes in the thermal conduction time constant. The edge computing chip locates the moment when the near-infrared light source is turned off in the temperature change sequence, extracts the cooling segment data after that moment, discretizes the cooling segment data, and fits it using a preset function model. The thermal conduction time constant is then calculated using the parameters obtained from the fitting.

[0025] Furthermore, the edge computing chip incorporates a pre-trained and quantized multi-dimensional feature fusion network. The edge computing chip inputs the calculated absorption peak position offset and thermal conduction time constant into this multi-dimensional feature fusion network. Specifically, the multi-dimensional feature fusion network adopts a lightweight fully connected network structure, adapted to the computing power and storage resources of the edge computing chip. The network's weight parameters are integer-quantized after training and stored in the edge computing chip's built-in flash memory. The multi-dimensional feature fusion network performs feature mapping and fusion processing on the input absorption peak position offset and thermal conduction time constant, outputting the diffusion uniformity index and compatibility level of the recycler in aged asphalt. The diffusion uniformity index is a continuous quantized value used to characterize the spatial distribution uniformity of the recycler in aged asphalt; the compatibility level is a discrete hierarchical label used to characterize the microscopic compatibility between the recycler and aged asphalt.

[0026] In this embodiment, an edge computing chip controls a near-infrared light source to emit a detection beam, simultaneously acquiring near-infrared diffuse reflectance spectral data and temperature change sequences of the recycled asphalt to be tested. The absorption peak position shift and thermal conduction time constant of the corresponding characteristic functional groups of the recycling agent are extracted. The diffusion uniformity index and compatibility level are output through a built-in multi-dimensional feature fusion network, realizing in-situ real-time detection of recycled asphalt performance. There is no need to transfer the sample to the laboratory for offline testing, preserving the true structure of the asphalt in the in-situ state of construction, and directly analyzing the microscopic diffusion and dissolution state of the recycling agent.

[0027] refer to Figure 2In a preferred embodiment, the edge computing chip generates a pulse modulation signal with a first timing sequence. This first timing pulse modulation signal includes two complementary pulse sub-signals with the same period and a phase difference of half a period. A preset guard interval is provided between the two pulse sub-signals to prevent overlap in their output. The edge computing chip transmits the first timing pulse modulation signal to the driving circuit of the near-infrared light source. The driving circuit drives the first and second light-emitting units of the near-infrared light source to work alternately according to the two pulse sub-signals, causing the near-infrared light source to alternately output a first center wavelength beam and a second center wavelength beam. Specifically, the first center wavelength beam corresponds to the first characteristic absorption wavenumber range of the characteristic functional groups in the regenerator, and the second center wavelength beam corresponds to the second characteristic absorption wavenumber range of the characteristic functional groups in the regenerator. The first and second characteristic absorption wavenumber ranges cover the absorption wavenumber range of all characteristic functional groups in the regenerator.

[0028] Furthermore, the edge computing chip synchronously transmits the pulse modulation signal of the first time sequence to the driving mechanism of the diffraction grating built into the micro spectrometer. The driving mechanism of the diffraction grating alternately switches according to the pulse modulation signal of the first time sequence, so that the beam splitting angle of the diffraction grating is synchronously matched with the center wavelength of the beam output by the near-infrared light source. The driving mechanism of the diffraction grating adopts a piezoelectric ceramic driving structure. The response speed of the piezoelectric ceramic driving structure is matched with the switching frequency of the pulse modulation signal. The edge computing chip outputs the corresponding driving voltage to the piezoelectric ceramic driving structure through the pre-stored correspondence between the driving voltage and the beam splitting angle, realizing the precise switching of the beam splitting angle of the diffraction grating, ensuring the synchronization of the beam splitting angle with the center wavelength of the beam output by the near-infrared light source, and avoiding the aliasing of spectral data in different bands. Specifically, when the near-infrared light source outputs a beam of the first central wavelength, the diffraction grating switches to the first beam-splitting angle, causing the photosensitive array of the miniature spectrometer to receive only the diffusely reflected light of the first band corresponding to the first central wavelength beam. When the near-infrared light source outputs a beam of the second central wavelength, the diffraction grating switches to the second beam-splitting angle, causing the photosensitive array of the miniature spectrometer to receive only the diffusely reflected light of the second band corresponding to the second central wavelength beam. The miniature spectrometer acquires spectral data of the first band during illumination by the first central wavelength beam and acquires spectral data of the second band during illumination by the second central wavelength beam. Both the first and second band spectral data contain light intensity values ​​of multiple sampling points within the corresponding wavenumber intervals.

[0029] Furthermore, the edge computing chip acquires the current temperature value from the internal ambient temperature sensor of the miniature spectrometer. This ambient temperature sensor is integrated onto the photosensitive array substrate of the miniature spectrometer and is used to collect the internal ambient temperature of the photosensitive array and diffraction grating in real time. Based on the acquired current temperature value, the edge computing chip performs a lookup operation in a pre-stored temperature drift compensation table to obtain the first compensation coefficient corresponding to the first band spectral data and the second compensation coefficient corresponding to the second band spectral data. Specifically, the temperature drift compensation table pre-stores the light intensity compensation coefficients corresponding to the first and second bands under different ambient temperatures. These compensation coefficients are obtained through a pre-calibrated temperature test, which is conducted in a temperature-controlled constant-temperature environment. Spectral data from a standard diffuse reflection white board is collected at different set temperatures, and the corresponding compensation coefficients at different temperatures are calculated based on the spectral intensity value at the standard ambient temperature. The edge computing chip divides the light intensity value corresponding to each wavenumber in the first band spectral data by the first compensation coefficient to complete the temperature drift correction of the first band spectral data; and divides the light intensity value corresponding to each wavenumber in the second band spectral data by the second compensation coefficient to complete the temperature drift correction of the second band spectral data.

[0030] Specifically, Table 2 is a comparison table of spectral compensation coefficients corresponding to different ambient temperatures. The compensation coefficients listed in the table are pre-stored in the built-in storage unit of the edge computing chip and are used for temperature drift correction of the spectral data collected by the micro spectrometer. For intermediate temperature values ​​not listed in the table, the edge computing chip calculates the corresponding compensation coefficients using linear interpolation.

[0031] Table 2 Comparison of Spectral Compensation Coefficients for Different Ambient Temperatures

[0032] Furthermore, the edge computing chip stitches the temperature-drift-corrected first-band spectral data with the second-band spectral data to generate complete near-infrared diffuse reflectance spectral data. Specifically, during the stitching process, the edge computing chip aligns the sampling points of the first-band and second-band spectral data using wavenumber as the abscissa. Before stitching, the edge computing chip performs wavelength calibration on the two bands' spectral data, using the characteristic absorption peaks of pre-stored standard substances as the wavelength calibration reference to calibrate the wavenumber coordinates of the two bands' spectral data, ensuring the accuracy of the wavenumber coordinates of the stitched spectral data and eliminating errors caused by wavelength drift. For the overlapping wavenumber interval between the two bands, the average light intensity value of the corresponding wavenumbers of the two bands is taken as the light intensity value of the stitched spectral data at that wavenumber, ultimately generating continuous near-infrared diffuse reflectance spectral data covering the preset complete wavenumber interval.

[0033] In this embodiment, the near-infrared light source is driven to alternately output beams with different center wavelengths by a pulse modulation signal in the first time sequence, and the diffraction grating is switched synchronously to achieve time-division acquisition of spectral data in a wide wavenumber range, thereby improving the signal-to-noise ratio and wavelength resolution of the spectral data. By collecting the ambient temperature inside the miniature spectrometer and looking up the corresponding compensation coefficient in the table, the spectral data is corrected, eliminating the drift interference of ambient temperature fluctuations on the spectral intensity values ​​and improving the accuracy of the spectral data.

[0034] refer to Figure 3 In a preferred embodiment, a high-density micro-thermistor array is arranged in a ring around the periphery of the near-infrared light source. The center of the ring coincides with the center of the light emission window of the near-infrared light source. The high-density micro-thermistor array includes multiple micro-thermistors uniformly distributed at equal angles. The detection surface of each micro-thermistor faces the detection surface of the recycled asphalt to be tested, and the detection surfaces of all micro-thermistors are on the same plane to ensure consistent adhesion to the surface of the recycled asphalt. Specifically, each micro-thermistor is a negative temperature coefficient ceramic semiconductor thermistor, whose resistance decreases as the temperature increases, to achieve high-sensitivity acquisition of the temperature of the micro-area of ​​the recycled asphalt to be tested.

[0035] Furthermore, each micro-thermistor in the high-density micro-thermistor array is connected to a corresponding Wheatstone bridge. Each Wheatstone bridge includes a micro-thermistor, a fixed resistor, a first matching resistor, and a second matching resistor. Specifically, the Wheatstone bridge adopts a four-arm structure: the first arm is the micro-thermistor, the second arm is the fixed resistor, the third arm is the first matching resistor, and the fourth arm is the second matching resistor. The micro-thermistor and the fixed resistor are connected in series to form a first series branch, and the first matching resistor and the second matching resistor are connected in series to form a second series branch. The first and second series branches are connected in parallel. The two input terminals of the Wheatstone bridge are connected to the output terminal of the constant current source and the ground terminal, respectively. The two output terminals of the Wheatstone bridge are the connection nodes of the micro-thermistor and the fixed resistor, and the connection nodes of the first matching resistor and the second matching resistor, respectively. The voltage difference between the two output terminals is the output voltage value of the Wheatstone bridge. The four-arm structure of the Wheatstone bridge adopts a symmetrical layout. The temperature coefficients of the first and second matching resistors are exactly the same as those of the micro thermistor. When the ambient temperature inside the smart sensor probe changes, the resistance values ​​of the micro thermistor and the reference thermistor will change by the same magnitude. The output voltage of the Wheatstone bridge is only related to the temperature change of the micro-area of ​​the recycled asphalt to be tested, and is independent of the ambient temperature change inside the probe, thus achieving complete suppression of common-mode interference from ambient temperature.

[0036] Furthermore, the first and second matching resistors are reference thermistors with the same temperature coefficient as the micro-thermistors. These reference thermistors are encapsulated in a heat-insulating cavity inside the smart sensor probe. A heat-insulating layer is provided between the heat-insulating cavity and the recycled asphalt to be tested, ensuring no heat conduction path between the reference thermistors and the recycled asphalt. The temperature of the reference thermistors is only consistent with the ambient temperature inside the smart sensor probe and is independent of the micro-area temperature of the recycled asphalt. The edge computing chip applies a constant current source excitation to the input of each Wheatstone bridge. The current value of the constant current source is output by the digital-to-analog converter built into the edge computing chip, ensuring the stability of the excitation. The edge computing chip's built-in storage unit pre-stores the resistance-temperature calibration curve corresponding to each micro-thermistor. The calibration curve is obtained through a pre-tested isothermal calibration test. At multiple preset calibration temperature points, the resistance values ​​of the micro-thermistors are collected, and the correspondence between the resistance value and the temperature value is obtained through polynomial fitting, ensuring the accuracy of temperature acquisition.

[0037] Furthermore, when the near-infrared light source is turned on, the edge computing chip synchronously acquires the output voltage value of each Wheatstone bridge through its built-in analog-to-digital converter and records it as the first output voltage value; when the near-infrared light source is turned off, it synchronously acquires the output voltage value of each Wheatstone bridge through the analog-to-digital converter and records it as the second output voltage value. During multiple consecutive cycles of alternating on and off of the near-infrared light source, the edge computing chip continuously acquires the first and second output voltage values, generating a voltage difference sequence corresponding to each micro-thermistor. Each element in the voltage difference sequence is the difference between the first and second output voltage values ​​within the same cycle. Based on a pre-stored resistance-temperature calibration curve, the edge computing chip converts the voltage difference sequence into a corresponding temperature value sequence, generating a temperature change sequence for the micro-region where each micro-thermistor is located.

[0038] Specifically, Table 3 is a table comparing the arrangement parameters of the micro thermistor array with the corresponding micro-area locations. The arrangement parameters of the micro thermistors listed in the table are used to define the ring distribution structure of the high-density micro thermistor array. Each micro thermistor corresponds to an independent micro-area on the surface of the recycled asphalt to be tested. By collecting temperature data from multiple micro-areas, the temperature distribution of the surface of the recycled asphalt to be tested can be fully covered, eliminating the random error of single-point acquisition.

[0039] Table 3. Comparison of the arrangement parameters of the micro-thermometer array with the corresponding micro-area positions.

[0040] In this embodiment, a high-density micro-thermometer array with a ring distribution is used to collect the temperature change sequence of multiple micro-regions of the recycled asphalt under test, ensuring the spatial coverage of temperature acquisition. By using a Wheatstone bridge structure containing a reference thermistor with the same temperature coefficient as the micro-thermometer, common-mode interference of ambient temperature fluctuations inside the smart sensor probe on temperature acquisition is eliminated, improving the acquisition accuracy of the temperature change sequence. Constant current source excitation and synchronous acquisition ensure the temporal consistency of temperature data and spectral data.

[0041] refer to Figure 4 In a preferred embodiment, the edge computing chip performs polynomial fitting on near-infrared diffuse reflectance spectral data to obtain a baseline curve. Specifically, the edge computing chip divides the near-infrared diffuse reflectance spectral data into multiple consecutive wavenumber windows according to a preset wavenumber interval. Each wavenumber window contains a fixed number of spectral sampling points, and adjacent wavenumber windows have a preset overlap area to avoid abrupt changes in the baseline curve at window junctions. For each wavenumber window, before performing polynomial fitting, the edge computing chip first filters the extreme points of the spectral data within the current wavenumber window, removing absorption peak extreme points and using only continuous spectral points without absorption peaks for baseline fitting. This avoids interference from absorption peaks with the baseline fitting results and ensures that the baseline curve accurately reflects the background drift of the spectrum. The edge computing chip sequentially uses quadratic, cubic, and quartic polynomials to fit the spectral data within the current wavenumber window, calculates the sum of squared residuals after fitting each polynomial, and selects the polynomial with the smallest sum of squared residuals as the target polynomial for the current wavenumber window. The edge computing chip calculates the baseline value corresponding to each wavenumber sampling point based on the target polynomial corresponding to each wavenumber window, and generates a continuous baseline curve.

[0042] The formula for calculating the sum of squared residuals is as follows:

[0043] in, For the sum of squared residuals, This represents the number of spectral sampling points within the current wavenumber window. For the first Wavenumber corresponding to each sampling point For the first The measured light intensity value corresponding to each sampling point. To fit the polynomial in wavenumber The calculated value at that location.

[0044] Furthermore, the edge computing chip subtracts the baseline value at the corresponding wavenumber in the baseline curve from the measured light intensity value at each sampling point in the near-infrared diffuse reflectance spectral data to obtain the corrected spectral data. The formula for baseline correction is as follows:

[0045] in, wave number The corrected spectral intensity value at that location, wave number The original near-infrared diffuse reflectance spectral intensity value at that location. wave number The baseline curve light intensity value at that location.

[0046] Furthermore, the edge computing chip determines a preset wavenumber interval for the corresponding regenerant characteristic functional groups in the calibration spectral data, searches for the minimum light intensity value within the preset wavenumber interval, and uses the wavenumber corresponding to the minimum point as the measured absorption peak position. Specifically, when searching for the minimum point within the preset wavenumber interval, the edge computing chip first performs a smoothing filter on the calibration spectral data, using a moving average filter with a preset window length to smooth the calibration spectral data, eliminating false minimum points caused by spectral noise, and ensuring the accuracy of absorption peak position extraction. The edge computing chip performs first derivative calculation on the calibration spectral data within the preset wavenumber interval. The wavenumber corresponding to the zero-crossing point where the first derivative changes from negative to positive is the wavenumber corresponding to the minimum point. When there are multiple minimum points within the preset wavenumber interval, the wavenumber corresponding to the minimum point with the smallest light intensity value is selected as the measured absorption peak position. The edge computing chip subtracts the measured absorption peak position from the pre-stored standard absorption peak position to obtain the absorption peak position offset. The formula for calculating the absorption peak position offset is as follows:

[0047] in, This represents the shift in the position of the absorption peak. To determine the wavenumber corresponding to the measured absorption peak position, The wavenumber corresponding to the pre-stored standard absorption peak position.

[0048] refer to Figure 5Furthermore, the edge computing chip extracts cooling segment data from the temperature change sequence starting from the moment the near-infrared light source is turned off, and discretizes this cooling segment data to obtain a discrete temperature point set. Specifically, the edge computing chip locates the timestamp corresponding to the moment the near-infrared light source turns off in the temperature change sequence, extracts all temperature data from that timestamp until the next moment the near-infrared light source turns on, and uses this as the cooling segment data. The cooling segment data is resampled at fixed time intervals to obtain a discrete temperature point set with equal time intervals, which includes the time sequence and the corresponding temperature sequence. Before fitting the cooling segment data, the edge computing chip first performs a stationarity check on the cooling segment data, calculates the first-order difference sequence of the cooling segment data, and when the fluctuation amplitude of the first-order difference sequence exceeds a preset threshold, the cooling segment data is determined to be invalid and discarded. Valid cooling segment data from adjacent periods is then used for fitting to ensure the reliability of the heat conduction time constant calculation. The edge computing chip uses a preset exponential decay function to perform least-squares fitting on the discrete temperature point set, and the reciprocal of the coefficient of the fitted exponential term is determined as the heat conduction time constant. The expression for the exponential decay function is as follows:

[0049] in, For time The corresponding micro-area temperature of the recycled asphalt to be tested, This represents the initial temperature rise at the moment the near-infrared light source is turned off. This is the time relative to the moment the near-infrared light source is turned off. The time constant of heat conduction. The ambient temperature is the temperature of the recycled asphalt to be tested.

[0050] To facilitate the solution of the least squares fit, the above exponential decay function is linearized, resulting in the linearized expression:

[0051] make , , Then the linearized expression simplifies to:

[0052] Edge computing chips use the least squares method to fit linearized discrete data points and solve for the coefficients. The value of the coefficient The reciprocal of the absolute value is used to determine the thermal conduction time constant of the corresponding micro-thermometer. The objective function for least-squares fitting is as follows:

[0053] in, The number of sampling points for the discrete temperature point set. For the first The linearized temperature value corresponding to each sampling point For the first The time corresponding to each sampling point.

[0054] Furthermore, the edge computing chip calculates the arithmetic mean of the thermal conduction time constants corresponding to each micro-thermometer in the high-density micro-thermometer array, and uses the calculated average value as the final thermal conduction time constant.

[0055] refer to Figure 6 Furthermore, the edge computing chip performs max-min normalization on the absorption peak position offset and the thermal conduction time constant to eliminate dimensional differences between different features. The formula for max-min normalization is as follows:

[0056] in, These are the standardized feature values. The original feature values ​​are to be standardized. The minimum value pre-stored for the corresponding feature. This is the pre-stored maximum value for the corresponding feature. For the absorption peak position offset, and These are the minimum and maximum values ​​of the pre-stored absorption peak position offset; for the thermal conduction time constant, and These are the minimum and maximum values ​​of the pre-stored heat conduction time constant.

[0057] Furthermore, the edge computing chip concatenates the standardized absorption peak position offset with the standardized thermal conduction time constant to generate a two-dimensional input vector. Specifically, the first element of the two-dimensional input vector is the standardized absorption peak position offset, and the second element is the standardized thermal conduction time constant. The edge computing chip inputs the two-dimensional input vector into the fully connected layer of the multi-dimensional feature fusion network pre-installed within the chip. The fully connected layer of the multi-dimensional feature fusion network consists of two layers: the first layer is the input layer, which receives the two-dimensional input vector and outputs an intermediate vector with a dimension of 4; the second layer is the output layer, which receives the intermediate vector and outputs an output vector with a dimension of 2. An activation function layer is placed between the two fully connected layers. The network structure is lightweight, and the total number of parameters is controlled within the storage resources of the edge computing chip, ensuring that the inference process can be completed within a preset time, meeting the needs of real-time detection. The fully connected layer performs matrix multiplication and bias addition operations on the two-dimensional input vector, with the following formulas:

[0058] in, This is the output vector of the fully connected layer. This is the weight matrix of the fully connected layer. Given a two-dimensional input vector, This is the bias vector of the fully connected layer. The dimension of the weight matrix matches the dimension of the two-dimensional input vector, so that the output vector of the fully connected layer contains two elements, corresponding to the intermediate mapping values ​​of diffusion uniformity and compatibility, respectively.

[0059] Furthermore, the output vector of the fully connected layer is input to the activation function layer, which uses the ReLU activation function to perform a non-linear mapping on each element of the output vector. The expression for the ReLU activation function is as follows:

[0060] in, The input values ​​for the activation function. This is the output value of the activation function.

[0061] The activation function layer outputs a first value and a second value. The edge computing chip inputs the first value into a pre-stored first piecewise linear mapping function, and the output value of the first piecewise linear mapping function is determined as the diffusion uniformity index. The expression of the first piecewise linear mapping function is as follows:

[0062] in, As an index of diffusion uniformity, The first value output by the activation function layer. , For the preset segmentation nodes, , , Let be the slope of each segment. , , The intercepts for each segment are given. The diffusion uniformity index is set within a predetermined continuous range, and its value is positively correlated with the diffusion uniformity of the recycling agent in aged asphalt.

[0063] Furthermore, the edge computing chip inputs the second numerical value into a pre-stored second piecewise linear mapping function, compares the output value of the second piecewise linear mapping function with multiple pre-stored compatibility level thresholds, and outputs the label corresponding to the compatibility level threshold that matches the comparison as the compatibility level output. The expression of the second piecewise linear mapping function is as follows:

[0064] in, For compatibility mapping values, The second value is the output of the activation function layer. , , For the preset segmentation nodes, , , , Let be the slope of each segment. , , , The intercepts for each segment are given.

[0065] Specifically, Table 4 is a compatibility level threshold and corresponding label comparison table. The compatibility level thresholds and corresponding labels listed in the table are pre-stored in the built-in storage unit of the edge computing chip. The edge computing chip compares the calculated compatibility mapping value with the interval in the table, determines the corresponding compatibility level label, and outputs it.

[0066] Table 4 Compatibility Level Thresholds and Corresponding Labels

[0067] Furthermore, the training process of the multi-dimensional feature fusion network is pre-completed on the server side. The training dataset contains synchronously collected data from multiple sets of recycled asphalt samples and corresponding gold standard labels. Specifically, each set of data in the training dataset includes near-infrared diffuse reflectance spectral data of the recycled asphalt sample, temperature change sequence, corresponding absorption peak position shift, and thermal conduction time constant. The gold standard labels include quantified values ​​of recycling agent diffusion uniformity and compatibility grading labels obtained through laboratory microscopic imaging equipment. During training, using the gold standard labels as the supervision target, the gradient descent algorithm is used to iteratively optimize the network's weight parameters and bias parameters until the error between the network's output value and the gold standard labels meets the preset requirements. After training, the network model undergoes integer quantization processing, is converted into a model format suitable for edge computing chip deployment, and stored in the built-in storage unit of the edge computing chip.

[0068] In this embodiment, the interference of spectral baseline drift is eliminated by automatically matching the polynomial with the minimum sum of squared residuals within the divided wavenumber window to generate the baseline curve, thus improving the accuracy of absorption peak position extraction. The thermal conductivity time constant is accurately calculated by performing least-squares fitting on the discrete temperature point set of the cooling section using an exponential decay function, which characterizes the thermal conductivity properties of recycled asphalt. The differences between different feature dimensions are eliminated by minimax normalization. Feature fusion is achieved through a lightweight fully connected network. The network output is converted into directly interpretable diffusion uniformity indicators and compatibility levels through a piecewise linear mapping function, adapting to the deployment requirements of edge computing chips and realizing in-situ quantitative evaluation of recycled asphalt performance.

Claims

1. A method for evaluating the performance of asphalt recycling based on intelligent sensors, characterized in that, The method is applied to an intelligent sensor probe that includes a near-infrared light source, a miniature spectrometer, a high-density micro-thermometer array, and an edge computing chip. The method includes: the edge computing chip controlling the near-infrared light source to emit a detection beam at a preset frequency. The miniature spectrometer collects near-infrared diffuse reflectance spectral data of the recycled asphalt to be tested, and the high-density micro-thermometer array synchronously collects the temperature change sequence of the micro-region of the recycled asphalt to be tested during the alternating on and off cycle of the near-infrared light source. The edge computing chip extracts the position offset of the absorption peak of the corresponding regenerant characteristic functional group in the near-infrared diffuse reflectance spectral data, and calculates the thermal conduction time constant in the temperature change sequence. The edge computing chip inputs the absorption peak position offset and the thermal conduction time constant into a multi-dimensional feature fusion network pre-set within the edge computing chip, and outputs the diffusion uniformity index and compatibility level of the regenerator in aged asphalt.

2. The method for evaluating the performance of asphalt recycling based on intelligent sensors according to claim 1, characterized in that, The edge computing chip controls the near-infrared light source to emit a detection beam at a preset frequency, including: the edge computing chip generates a pulse modulation signal with a first timing sequence, and drives the near-infrared light source to alternately output a first center wavelength beam and a second center wavelength beam according to the pulse modulation signal with the first timing sequence; The miniature spectrometer acquires near-infrared diffuse reflectance spectral data of the recycled asphalt to be tested, including: the diffraction grating built into the miniature spectrometer alternately switches according to the pulse modulation signal of the first time sequence; the miniature spectrometer acquires first band spectral data during the illumination of the first center wavelength beam and acquires second band spectral data during the illumination of the second center wavelength beam; and the edge computing chip stitches the first band spectral data and the second band spectral data into the near-infrared diffuse reflectance spectral data.

3. The method for evaluating the performance of asphalt recycling based on intelligent sensors according to claim 1, characterized in that, The high-density micro-thermistor array synchronously acquires the temperature change sequence of the micro-region of the recycled asphalt under test during the alternating on and off cycle of the near-infrared light source. This includes: each micro-thermistor in the high-density micro-thermistor array is connected to a corresponding Wheatstone bridge; the edge computing chip records the first output voltage value of the Wheatstone bridge when the near-infrared light source is turned on, and records the second output voltage value of the Wheatstone bridge when the near-infrared light source is turned off; the edge computing chip generates the temperature change sequence of the micro-region where each micro-thermistor is located based on the difference sequence between the first output voltage value and the second output voltage value; wherein the high-density micro-thermistor array is distributed in a ring around the periphery of the near-infrared light source.

4. The method for evaluating the performance of asphalt recycling based on intelligent sensors according to claim 1, characterized in that, The edge computing chip extracts the absorption peak position offset of the corresponding regenerant characteristic functional group in the near-infrared diffuse reflectance spectral data, including: the edge computing chip performs polynomial fitting on the near-infrared diffuse reflectance spectral data to obtain a baseline curve, subtracts the baseline curve from the near-infrared diffuse reflectance spectral data to obtain corrected spectral data, the edge computing chip determines a preset wavenumber interval corresponding to the regenerant characteristic functional group in the corrected spectral data, finds the wavenumber corresponding to the minimum point in the preset wavenumber interval as the measured absorption peak position, and subtracts the measured absorption peak position from the pre-stored standard absorption peak position to obtain the absorption peak position offset.

5. The method for evaluating the performance of asphalt recycling based on intelligent sensors according to claim 1, characterized in that, The edge computing chip calculates the heat conduction time constant in the temperature change sequence, including: the edge computing chip extracts cooling segment data from the temperature change sequence starting from the moment the near-infrared light source is turned off; the cooling segment data is discretized to obtain a discrete temperature point set; the edge computing chip performs least squares fitting on the discrete temperature point set using a preset exponential decay function; the reciprocal of the coefficient of the fitted exponential term is determined as the heat conduction time constant; and the edge computing chip calculates the average value of the heat conduction time constant corresponding to each micro-thermistor in the high-density micro-thermistor array as the final heat conduction time constant.

6. The method for evaluating the performance of asphalt recycling based on intelligent sensors according to claim 1, characterized in that, The edge computing chip inputs the absorption peak position offset and the thermal conduction time constant into a multi-dimensional feature fusion network pre-installed within the edge computing chip, including: the edge computing chip performing maximum and minimum normalization processing on the absorption peak position offset and the thermal conduction time constant respectively; concatenating the normalized absorption peak position offset and the normalized thermal conduction time constant into a two-dimensional input vector; inputting the two-dimensional input vector into a fully connected layer of the multi-dimensional feature fusion network; the fully connected layer performing matrix multiplication and bias addition operations on the two-dimensional input vector and then inputting it into an activation function layer; and the activation function layer outputting the diffusion uniformity index and the compatibility level.

7. The method for evaluating the performance of asphalt recycling based on intelligent sensors according to claim 2, characterized in that, Before the edge computing chip stitches the first band spectral data and the second band spectral data into the near-infrared diffuse reflectance spectral data, the method further includes: the edge computing chip obtaining the current temperature value of the ambient temperature sensor inside the micro spectrometer, looking up the current temperature value in a pre-stored temperature drift compensation table to obtain the first compensation coefficient corresponding to the first band spectral data and the second compensation coefficient corresponding to the second band spectral data, the edge computing chip dividing the light intensity value of each wavelength in the first band spectral data by the first compensation coefficient, and dividing the light intensity value of each wavelength in the second band spectral data by the second compensation coefficient.

8. The method for evaluating the performance of asphalt recycling based on intelligent sensors according to claim 3, characterized in that, Each micro-thermistor in the high-density micro-thermistor array is connected to a corresponding Wheatstone bridge, including: the Wheatstone bridge includes the micro-thermistor, a fixed resistor, a first matching resistor, and a second matching resistor, the first matching resistor and the second matching resistor are reference thermistors with the same temperature coefficient as the micro-thermistor, the reference thermistor is encapsulated in a heat-insulating cavity inside the smart sensor probe to isolate the recycled asphalt to be tested, the edge computing chip applies a constant current source excitation to the Wheatstone bridge, and the first output voltage value and the second output voltage value are the voltage difference between the two ends of the diagonal of the Wheatstone bridge.

9. The method for evaluating the performance of asphalt recycling based on intelligent sensors according to claim 4, characterized in that, The edge computing chip performs polynomial fitting on the near-infrared diffuse reflectance spectral data to obtain a baseline curve, including: the edge computing chip divides the near-infrared diffuse reflectance spectral data into multiple consecutive wavenumber windows; for each wavenumber window, the edge computing chip sequentially uses a quadratic polynomial, a cubic polynomial, and a quartic polynomial to fit the spectral data within the current wavenumber window; calculates the sum of squared residuals after fitting each polynomial; selects the polynomial with the smallest sum of squared residuals as the target polynomial for the current wavenumber window; and generates a continuous baseline curve based on the target polynomial corresponding to each wavenumber window.

10. The method for evaluating the performance of asphalt recycling based on intelligent sensors according to claim 6, characterized in that, The activation function layer outputs the diffusion uniformity index and the compatibility level, including: the activation function layer outputs a first value and a second value; the edge computing chip inputs the first value into a pre-stored first piecewise linear mapping function, determines the output value of the first piecewise linear mapping function as the diffusion uniformity index; the edge computing chip inputs the second value into a pre-stored second piecewise linear mapping function, compares the output value of the second piecewise linear mapping function with multiple pre-stored compatibility level thresholds, and outputs the label corresponding to the compatibility level threshold that matches the comparison as the compatibility level.

Images