Vehicle-mounted airport runway pollutant detection method based on ultrasonic technology

Abstract

The invention relates to the field of aviation safety and airport runway detection, and provides a vehicle-mounted airport runway pollutant detection method based on an ultrasonic technology. The method comprises a vehicle-mounted ultrasonic sensor design technology, a pollutant type identification technology and a depth detection technology. The sensor is composed of an ultrasonic probe, a lifting motor, a rotating motor, a waterproof cover, a clamp and an aviation plug. The pollutant type identification step comprises the steps of obtaining a pollutant echo signal data set, extracting waveform characteristic parameters (such as amplitude, distortion degree and signal-to-noise ratio), constructing a CNN deep learning model, performing neural network training and outputting an identification result. The pollutant depth detection technology is based on ultrasonic echo parameter estimation, and comprises the following steps: extracting an echo signal envelope line, retrieving feature points, constructing a three-objective function (based on a feature point difference value and a loss function value), solving a model parameter optimal solution by adopting an OMOPSO algorithm, and calculating the pollutant depth.

Description

Technical Field

[0001] The invention relates to an airport runway ice and snow water information detection technology, including ice and snow water depth detection and type identification, and relates to ultrasonic sensors and echo signal processing technology. Background Art

[0002] In the aviation field, the safety performance of the runway is crucial to ensure the normal takeoff and landing of aircraft. Especially in severe weather conditions, such as when the runway surface is covered with ice, snow or water, its anti-skid performance will be greatly reduced, thereby increasing the risk of tire slippage and causing the aircraft to run off the runway. This risk not only threatens the lives of passengers and crew members, but may also have a significant impact on airport operations. Therefore, accurate monitoring of the depth of contaminants on the runway surface has become an important task in airport safety management.

[0003] According to Circular 355 issued by ICAO, pollutants include water, ice, wet snow, dry snow, slush and other coverings, and the depth of pollutants on the runway surface needs to be accurately measured. The detection of the depth of runway pollutants must reach millimeter-level accuracy. This requirement is based on an in-depth study of the relationship between the anti-skid performance of the runway surface and the depth of pollutants. When the depth of pollutants reaches 3 mm, the friction coefficient of the runway will be significantly reduced, whether it is dry snow / wet snow or slush / water. In addition, when the aircraft is taxiing at high speed, the fluid pollutants will generate hydrodynamic pressure on the aircraft tires, which will cause the tires to lose contact with the runway surface. At this time, the braking effect of the tires is greatly reduced, and the risk of the aircraft running off the runway increases.

[0004] In order to meet this challenge, it is necessary to develop runway surface pollutant detection technology. The detection of ice / snow / water on the runway surface is mainly divided into two types: contact and non-contact. The contact method mainly uses sensors buried on the runway surface to monitor the ice / snow / water on the runway in real time. The non-contact method mainly uses remote sensing sensors to scan, measure distance and take pictures of runway pollutants. Common remote sensing sensors include laser, ultrasonic, infrared sensors, and optical / infrared cameras. These remote sensing sensors can be installed on the shoulders of the runway (mainly installed on the weather station) or on the detection vehicle. As a point detection, the embedded sensor has a small detection area, and the exposed runway surface of the sensor is easily damaged and has a low lifespan. In addition, it is difficult for the airport to meet the runway surface condition assessment work in terms of detection result accuracy, monitoring cost, and maintenance cost. Although the meteorological monitoring station is powerful and can detect diverse information, its detection range is only near the installation location and it is difficult to cover the entire airport. If multiple runway sections are to be detected, multiple meteorological stations need to be installed. For example, Wellington Airport in New Zealand has installed RWS (located in the take-off and landing areas of the two sections of the runway and the centerline of the runway). However, the deployment of RWS at multiple points will greatly increase the monitoring cost, and due to the limited height requirements of the airport itself, the low installation height of the remote sensing sensor will also reduce the detection range. The detection vehicle has become the first choice for many airports to detect pavement conditions due to its high detection efficiency and wide detection range.

[0005] At present, vehicle-mounted sensors that detect ice and snow on the surface mostly use optical technology, including laser, spectroscopy, and ultrasonic remote sensing measurement technology. For example, the MD30 sensor developed by Finland's Vaisala uses laser technology, the MARWIS developed by Germany's Lufft and the RCM411 sensor by Finland's Teconer all use spectroscopy technology. Although these methods can obtain information on the type and depth of road pollutants, their detection range is limited, and the depth measurement results for transparent media such as snow slurry are inaccurate, and it is difficult to achieve millimeter-level accuracy. In addition, ambient light will also cause certain interference to infrared and spectral sensors, resulting in poor reliability of the results.

[0006] Ultrasonic detection of ice and snow is not affected by transparent media or light, and can achieve millimeter-level or higher accuracy through appropriate signal processing. At the same time, based on the test, the echo signals of different pollutants are obtained, and the waveform characteristics are obviously different. Therefore, it is potential to identify ice, water, wet snow, dry snow, and slurry based on the echo signal characteristics. This feature makes ultrasonic sensors an ideal choice for measuring the depth of pollutants on detection vehicles. Ultrasonic sensors can not only meet the requirements of millimeter-level accuracy, but also adapt to various harsh runway conditions, providing strong support for runway surface evaluation, friction / anti-skid performance analysis, and management / maintenance efficiency. Therefore, ultrasonic sensors have broad application prospects in the field of runway pollutant depth detection. Summary of the invention

[0007] In order to solve the above technical problems, the present application aims to design a runway surface pollutant detection technology, including a vehicle-mounted ultrasonic sensor, a pollutant type identification technology and a pollutant depth detection technology.

[0008] The present application achieves the above effects through the following technical solutions: a vehicle-mounted airport runway pollutant detection method based on ultrasonic technology, the method is based on the following detection system:

[0009] Design of vehicle-mounted ultrasonic sensor;

[0010] The vehicle-mounted ultrasonic sensor is installed on the vehicle, and includes an ultrasonic air probe, which adopts a single-transmitting and single-receiving air-coupled probe. The probe is embedded in a PCB circuit board, and the PCB board is connected to the rotating motor;

[0011] The vehicle-mounted ultrasonic sensor also includes a rotating motor and a lifting motor. The rotating motor controls the rotation direction and speed of the ultrasonic air probe by changing the polarity and magnitude of the voltage, and the lifting motor realizes the lifting and lowering of the ultrasonic air probe through a screw rod.

[0012] The vehicle-mounted ultrasonic sensor also includes a waterproof cover, the inner wall of which is fixed to the lifting motor, the outside of which is connected to the fixed rod, an aviation plug is installed on the side, and a hole is opened at the bottom for the probe to receive the echo signal.

[0013] The vehicle-mounted ultrasonic sensor is installed on the vehicle through a clamp, which consists of a detachable steel connecting rod, a fixed metal plate, and hand screws. The steel rod can be spliced ​​and disassembled in sections to change the length; the fixed metal plate is used to fix the waterproof cover and the connecting rod.

[0014] The vehicle-mounted ultrasonic sensor is installed on the vehicle, including a single-transmitting and single-receiving air-coupled ultrasonic air probe. The sensor is equipped with a lifting and rotating motor to control the rotating probe to detect the height and angle. The lifting motor realizes the lifting of the probe through a screw rod. The probe is embedded in a PCB circuit board and connected to the rotating motor. A waterproof cover is arranged outside the probe and the motor for waterproof and dustproof, and an aviation plug is arranged outside the waterproof cover for signal transmission.

[0015] The sensor is installed on the vehicle through a fixture consisting of a detachable steel connecting rod, a fixed metal plate and hand screws. The steel rod can be adjusted in length to suit different installation requirements.

[0016] Furthermore, the pollutant detection information includes pollutant type and depth information. The vehicle-mounted ultrasonic sensor first performs type identification and then performs depth calculation. The type identification includes obtaining echo signal samples of various pollutants, extracting echo signal waveform characteristic parameters, constructing a CNN deep learning model, adjusting CNN model parameters, and outputting pollutant type identification results.

[0017] The echo signal waveform parameters include the amplitude, distortion degree and signal-to-noise ratio of the echo signal.

[0018] Further, the echo signal amplitude is expressed as:

[0019] A=max[S p (i)]

[0020] Among them, A represents the amplitude of the echo signal, S p (t) represents the maximum voltage amplitude in the obtained echo signal time series point, i is the time series point of the echo signal;

[0021] The degree of distortion is represented by cosine similarity, which is expressed as:

[0022]

[0023] Among them, S p ,S d Represents the actual ultrasonic echo signal, S d Represents the echo signal of the standard mixed exponential model, and N represents the total number of signal sampling points

[0024] The echo signal of the mixed exponential model is specifically:

[0025]

[0026] Among them, t i is a time series point, A0 is the amplitude of the signal, m,T is the intrinsic parameter of the ultrasonic probe, which is used to characterize the waveform skewness and kurtosis of the signal, τ is the offset of the starting point, f0 is the center frequency, and θ is the initial phase;

[0027] The signal-to-noise ratio of the echo signal is expressed as:

[0028]

[0029] Among them, P s and P n represent the signal power and noise power respectively.

[0030] Furthermore, the CNN deep learning model includes a convolutional layer, an input layer, a hidden layer, a fully connected layer and an output layer. The convolutional layer is used to extract local features in the echo signal, and the fully connected layer is used to map the extracted features to the pollutant category; the neural network input is the echo signal waveform parameters, and the output is the type of pollutant.

[0031] Furthermore, the depth detection method includes echo signal envelope extraction, echo signal feature point retrieval, construction of a multi-objective function, and use of the OMOPSO method to obtain the optimal solution of the multi-objective function and pollutant depth calculation.

[0032] Furthermore, the echo signal envelope extraction process uses wavelet denoising, Hilbert transform and cubic spline curve smoothing methods to filter the echo signal to obtain a smooth signal, specifically:

[0033] The process of wavelet denoising includes using Nakagami wavelet basis function, selecting wavelet coefficients, decomposition scale and threshold, using soft threshold method to remove signal noise, and finally reconstructing the signal;

[0034] The Hilbert transform is:

[0035] z(i)=H{S p (i)}

[0036] R 0p (i)=|z(i)|

[0037] Among them, H{S p (i)} is the Hilbert operation of the echo signal, z(i) is a complex signal, |z(i)| is the modulus of the complex signal;

[0038] The cubic spline interpolation method is:

[0039]

[0040] Among them, R p (i) is the interpolation curve, that is, the smoothed envelope, and s(i) is the cubic spline basis function.

[0041] Among them, the basis function s(i) can be expressed as:

[0042] s(i)=a i +b i (xx i )+c i (xx i )2+d i (xx i )

[0043] Among them, the coefficient a i , b i , c i , d i It is obtained through interpolation conditions, curve continuity conditions, and continuity conditions of the first-order derivative and second-order derivative of the curve.

[0044] Furthermore, the echo signal feature points include:

[0045] The first characteristic point is the peak point of the echo signal envelope function, which is expressed as:

[0046] Result[i1]suject to R p (i1)=max(R p (i))

[0047] Among them, Result[i1] represents the search result of the first feature point in the time series;

[0048] The second characteristic point is the peak point of the first-order derivative of the echo signal envelope function, which is expressed as:

[0049]

[0050] Among them, Result[i2] represents the search result of the second feature point in the time series. Indicates the envelope R p (i) The first derivative of

[0051] Furthermore, the multi-objective function includes:

[0052] The first objective function is to minimize the loss function value between the actual echo signal and the fitted exponential function signal, and its expression is:

[0053]

[0054] Where loss(x) is the loss function when the echo signal parameters are x = [A, m, T, τ], S f is the fitting function;

[0055] The second objective function is to minimize the difference between the first characteristic point of the actual echo signal and the fitted exponential function signal in the time series, and its expression is:

[0056]

[0057] in, Retrieve the difference between the time series point of the fitting function signal and the first characteristic point of the actual signal under the echo signal parameter x = [A, m, T, τ], R f is the envelope function of the fitting function signal;

[0058] The third objective function, wherein the difference between the retrieval results of the actual echo signal and the second characteristic point of the fitted exponential function signal in the time series is the smallest, is expressed as:

[0059]

[0060] in, Retrieving a difference value for a time series point of a second characteristic point of a fitting function signal and an actual signal under the echo signal parameter x=[A, m, T, τ];

[0061] The final optimization goal based on the three objective functions is:

[0062] minL(x)=[L(x)1,L(x)2,L(x)3]

[0063]

[0064] Among them, L(x)1, L(x)2, and L(x)3 are three objective functions.

[0065] Furthermore, the steps of obtaining the optimal solution of the multi-objective function using the OMOPSO method include: initializing particle position and velocity parameters, updating and mutating particle position and velocity, storing solutions of non-dominated particles, judging output conditions, and outputting a Pareto optimal solution set;

[0066] Among them, the speed of each particle is expressed as:

[0067]

[0068] Among them, C1 and C2 of OMOPSO are control and gbest t The acceleration constant that affects the particle velocity, C1 and C2 are random numbers in [0,2], ω is a random number, r1 and r2 are used to increase the randomness of the search process and are random numbers in [0,1]; w t is the inertia weight of particle i and controls the trade-off between global history and local history; and is the position and velocity of particle i at time t;

[0069] After obtaining the velocity of each particle, the position of particle i is updated, and the updated particle position is expressed as:

[0070]

[0071] The Archive is a solution set of non-dominated solutions of the multi-objective function. The size of Archive is determined by ∈, and ∈ satisfies the following formula:

[0072]

[0073] The size of the Archive is determined by ∈-dominance, where the value of ∈ limits the size of the Archive, the value of ∈ is defined by the user, and the decision vector x1 is ∈-dominance x2;

[0074] The crowding degree is the density of the solution of the multi-objective function in the search space, which is expressed as follows:

[0075]

[0076] Among them, i d represents the crowding degree of the i-th non-dominated particle in Archive, represents the jth objective function value at point i+1, represents the j-th objective function value at point i-1, and m is the number of objective functions.

[0077] The beneficial effects of this application are as follows: the vehicle-mounted pollutant information detection method based on ultrasonic technology proposed in this patent can realize the accurate identification of pollutants, including water / ice / wet snow / slush / dry snow types and precise calculation of depth. At the same time, the vehicle-mounted sensor can dynamically obtain runway surface pollutant information to improve detection efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] Figure 1 The three-view diagram of the vehicle-mounted ultrasonic sensor of the present invention;

[0079] Figure 2 A three-dimensional diagram of the vehicle-mounted ultrasonic sensor of the present invention;

[0080] Figure 3 It is an ultrasonic echo signal diagram of ultrasonic water / ice / snow mentioned in the implementation case of the present invention;

[0081] Figure 4 This is a flow chart of the pollutant type identification method based on CNN of the present invention;

[0082] Figure 5 This is a flow chart of the depth detection method based on the OMOPSO method of the present invention;

[0083] Figure 6 This is a program diagram for obtaining the optimal waveform parameters based on the OMOPSO algorithm described in the present invention. DETAILED DESCRIPTION

[0084] The present invention relates to a runway surface pollutant detection technology, and its specific implementation method is as follows:

[0085] like Figure 1 and Figure 2 As shown, this embodiment provides a vehicle-mounted airport runway pollutant detection method based on ultrasonic technology. The solution is based on a vehicle-mounted ultrasonic sensor. The invented vehicle-mounted ultrasonic sensor includes an excitation probe 1, a receiving probe 2, a lifting motor 3, a rotating motor 4, a waterproof cover 5, a clamp 6 and an aviation plug 7. The rotating motor is connected to a screw rod 8, and the clamp includes a connecting rod 9 for connecting the sensor and the clamp, and a hand screw 10 for adjusting the clamp.

[0086] like Figure 1 and Figure 2 As shown, the present invention adopts a 40kHz single-transmit single-receive air-coupled probe 1, which is designed with two probes with a horizontal distance of 15 mm to ensure that the sound wave generating surface is on the same horizontal plane. The probe surface is designed to be waterproof to adapt to the outdoor environment. The maximum input voltage of the probe is 15Vpp, and the working voltage in the present invention is 10Vpp, which is suitable for a temperature range of -20°C to 70°C. The probe is fixed on a circuit board, which is treated with epoxy resin for waterproofing and dustproofing, and is connected to a rotating motor to achieve precise positioning.

[0087] like Figure 1 and Figure 2 As shown, the present invention designs a lifting motor 2 and a rotating motor 3 to adjust the detection angle and height of the sensor. The circuit board of the ultrasonic probe is mounted on a rotating plate, and the rotating plate is connected and fixed to the rotating shaft in the motor by screws. The rotation direction and speed of the motor can be controlled by changing the polarity and magnitude of the voltage, and the rotation angle range is -45° to 45°. The rotating motor is mounted on a sliding base, and the sliding base is mounted on a screw rod, which is connected to the rotating shaft of the motor. The horizontal position of the probe is lifted and lowered by rotating the screw rod, and the lifting range is 10 cm.

[0088] like Figure 1 and Figure 2As shown, the waterproof cover 4 of the present invention is designed to have a height of 15 cm, a width of 10 cm and a length of 15 cm respectively, the inner wall is fixed to the lifting motor by screws, and the outside is connected to the fixed rod. The lower part of the waterproof cover is opened for generating and receiving ultrasonic signals, and an aviation plug is installed on the side for circuit connection and motor control.

[0089] like Figure 1 and Figure 2 As shown, the fixture consists of a detachable steel rod, a fixed metal plate, and a thumb screw. The ultrasonic detection device is connected to the fixed metal plate through a waterproof cover and fixed to the head end of the steel rod. A thumb screw is installed at the tail end of the fixture to fix the ultrasonic detection device on the vehicle. The steel rod can be arbitrarily changed in length by segmented splicing and disassembly to meet the installation requirements of different vehicles.

[0090] The vehicle-mounted ultrasonic sensor comprises:

[0091] Ultrasonic air probe, position and orientation control motor, waterproof cover, clamp, aviation plug.

[0092] The ultrasonic air probe is specifically a 40khz single-transmit and single-receive air-coupled probe, with a maximum input voltage of 15Vpp, an input voltage of 10Vpp, and an operating temperature of -20℃~70℃. The horizontal distance between the two probes is 15mm, and the sound wave generating surfaces of the two probes are on the same horizontal plane. At the same time, the probe surface is resistant to water drops, that is, splashing water during the detection process has no effect on the probe measurement work. The ultrasonic probe is fixed on the circuit board, the circuit board is treated with epoxy resin for waterproofing and dustproofing, and the circuit board is connected to the rotating motor.

[0093] The motor is specifically: Considering that in actual detection, the ultrasonic detection device is fixed at different positions on different detection vehicles, when the height and angle of the ultrasonic probe from the ground deviate from the preset position, the lifting motor and the rotating motor can be used to control the detection angle and detection height of the sensor. The circuit board of the ultrasonic probe is installed on the rotating plate, which is connected and fixed to the rotating shaft in the motor by screws, and the rotation direction and speed of the motor are controlled by changing the polarity and magnitude of the voltage, such as Figure 5 As shown in the figure, with the horizontal plane as 0°, the probe rotation angle range is -45° to 45°. The rotating motor is installed on the sliding base, and the sliding base is installed on the screw rod. The screw rod used is a spiral metal rod and is connected to the motor shaft. The thread on the screw rod, that is, the thread pattern, matches the nut. When the motor rotates, the thread guides the nut to move along the axial direction of the screw rod. The horizontal position of the probe is raised and lowered by rotating the screw rod, as shown in the figure. Figure 6 As shown, the length of the screw rod is 15 cm and the maximum range of the probe lifting height is 10 cm.

[0094] The waterproof cover is specifically as follows: the height of the waterproof cover is 15 cm, and the width and length are 10 cm and 15 cm. The inner wall of the waterproof cover is fixed to the lifting motor by screws, and the outside of the waterproof cover is connected to the fixed rod by screws. The lower part of the waterproof cover is opened for ultrasonic signal generation and reception, and the side of the waterproof cover is installed with an aviation plug. The jacks are used for ultrasonic probe circuit connection, lifting motor control and rotating motor control respectively. The bottom of the waterproof cover is opened so that the probe receives the echo signal reflected from the surface of the pollutant. The size of the opening is 8 cm × 8 cm.

[0095] The clamp is specifically composed of a detachable steel rod, a fixed metal plate, and a hand screw. The ultrasonic detection device is connected to the fixed metal plate through a waterproof cover and fixed to the head end of the steel rod by screws. The hand screw is installed at the tail end of the clamp to fix the ultrasonic detection device on the vehicle. The length of the steel rod can be arbitrarily changed by segmented splicing and disassembly.

[0096] The aviation plugs are specifically: three aviation plugs are all M12 aviation plugs, and the aviation plugs are connected to the positive and negative leads of the ultrasonic sensor and the wires of the lifting and rotating motors in the waterproof cover.

[0097] The pollutant type identification technology includes the acquisition of various pollutant echo signal data sets, the extraction of waveform feature parameters of the echo signals, the construction of a CNN deep learning model, neural network training and the output of pollutant type identification results.

[0098] like Figure 1 and Figure 2 As shown, the three aviation plugs used in the present invention are all of M12 type, which are connected to the positive and negative leads of the ultrasonic sensor and the wires of the lifting and rotating motors in the waterproof cover, ensuring the reliability and stability of the electrical connection.

[0099] The acquisition of the echo signal data set of various pollutants is specifically as follows: before performing the pollutant type identification task, first collect and organize the echo signal data set containing various pollutant types such as water, ice, wet snow, dry snow, and slush. To ensure the generalization ability and recognition accuracy of the model. The data is acquired by collecting the echo signals of winter snow samples and collecting the echo signals of different pollutants by artificial snowmaking in a low-temperature environment laboratory.

[0100] Extracting the waveform characteristic parameters of the echo signal is one of the key steps in identifying the type of pollutant. The echo signal characteristic parameters include: the amplitude of the echo signal, the degree of distortion and the signal-to-noise ratio. The extraction of these characteristic parameters is helpful for the subsequent training and identification of the deep learning model.

[0101] The present invention uses a deep learning method based on the echo signal characteristics of different types of pollutants to identify the types of pollutants. The identification method is as follows: Figure 4 As shown, it includes the following five steps.

[0102] Step 1: Obtain echo signal samples of various pollutants, including water, ice, wet snow, dry snow, and slurry echo signal samples.

[0103] The specific implementation is as follows: Figure 3 As shown, water, ice, wet snow, dry snow, and slush echo signals are collected and sorted to ensure the generalization ability and recognition accuracy of the model. Data is obtained by collecting echo signals of winter snow samples and collecting echo signals of different pollutants through artificial snowmaking in a low-temperature environment laboratory.

[0104] Step 2: Extract echo signal waveform characteristic parameters.

[0105] The specific implementation is as follows: Figure 4 As shown in the process, the waveform characteristic parameters of the echo signal are extracted, including the amplitude, distortion degree and signal-to-noise ratio of the echo signal. The extraction of these characteristic parameters is helpful for the subsequent training and recognition of the deep learning model.

[0106] The echo signal amplitude is expressed as:

[0107] A=max[S p (i)]

[0108] Among them, A represents the amplitude of the echo signal, S p (t) represents the maximum voltage amplitude in the acquired echo signal time series point, and i is the time series point of the echo signal.

[0109] The degree of distortion of the signal shown is expressed by cosine similarity, which is expressed as:

[0110]

[0111] Among them, S p ,S d Represent the actual ultrasonic echo signal and the echo signal of the standard mixed exponential model respectively

[0112] The echo signal of the mixed exponential model is specifically:

[0113]

[0114] Among them, t i is a time series point, A0 is the amplitude of the signal, m,T is the intrinsic parameter of the ultrasonic probe, which is used to characterize the waveform skewness and kurtosis of the signal, τ is the offset of the starting point, f0 is the center frequency, and θ is the initial phase.

[0115] The signal-to-noise ratio of the echo signal is expressed as:

[0116]

[0117] Among them, P s and P n represent the signal power and noise power respectively.

[0118] Step 3: Build a CNN deep learning model.

[0119] The specific implementation is as follows: Figure 4 As shown in the flow chart, a convolutional neural network is used to process and analyze echo signal data. The CNN model consists of an input layer, a hidden layer, a fully connected layer, and an output layer. The convolutional layer is used to extract local features in the echo signal, and the fully connected layer is used to map the extracted features to the pollutant category. The neural network input is the three characteristic parameters of the ultrasonic echo signal, namely the amplitude A, the echo distortion degree C s The output is the type of pollutant. The activation function used by the hidden layer neurons is the sigmoid function. There are three hidden layers, and the number of neurons in each layer is 20.

[0120] Step 4: Adjust CNN model parameters

[0121] The specific implementation is as follows: Figure 4 As shown in the process, the CNN model is trained and optimized using the extracted echo signal feature parameters and known pollutant category labels. During the training process, the model parameters are continuously adjusted to gradually improve the model's fit to the training data while avoiding overfitting. After the training is completed, the model will have the ability to identify pollutant types for unknown echo signals.

[0122] Step 5: Output the results

[0123] The specific implementation is as follows: Figure 4 As shown in the process, after the model training is completed, the echo signal to be identified is input into the model, and the final recognition result is obtained through the forward propagation calculation of the model.

[0124] Since the echo signal waveform obtained by the ultrasonic sensor is a mixed exponential waveform, the present invention analyzes the waveform parameters of the echo signal including x=[A, m, T, τ], and obtains the waveform parameters by obtaining the waveform rising edge of the actual echo signal that is closest to the echo rising edge of the standard mixed exponential model, wherein τ is the offset of the starting point, which is used to calculate the final time of flight (ToF) for calculating the depth of the pollutant.

[0125] The depth detection method described in the present invention includes five steps: echo signal envelope extraction, echo signal feature point retrieval, construction of a multi-objective function, use of the OMOPSO method to obtain the optimal solution of the multi-objective function and calculation of the pollutant depth.

[0126] Step 1: Extract the echo signal envelope

[0127] Specifically, it includes wavelet denoising of the echo signal and extracting the envelope function R using Hilbert transform 0p , and smooth the envelope function by cubic spline interpolation to obtain the final smooth envelope function R p .

[0128] The specific implementation is as follows: Figure 5 As shown in the flowchart in, the process of wavelet denoising includes using Nakagami wavelet basis function, selecting appropriate wavelet coefficients, decomposition scale and threshold, using soft threshold method to remove signal noise, and finally reconstructing the signal.

[0129] The Hilbert method for extracting the envelope is:

[0130] z(i)=H{S p (i)}

[0131] R 0p (i)=|z(i)|

[0132] Among them, H{S p (i)} is the Hilbert operation performed on the echo signal, z(i) is a complex signal, and |z(i)| is the modulus of the complex signal.

[0133] The cubic spline interpolation method is:

[0134]

[0135] Among them, R p (i) is the interpolation curve, that is, the smoothed envelope, and s(i) is the cubic spline basis function.

[0136] Among them, the basis function s(i) can be expressed as:

[0137] s(i)=a i +b i (xx i )+c i (xx i )2+d i (xx i )

[0138] Among them, the coefficient a i , b i , ci , d i It is obtained through interpolation conditions, curve continuity conditions, and continuity conditions of the first-order derivative and second-order derivative of the curve.

[0139] Step 2: Retrieve echo signal feature points.

[0140] The specific implementation is as follows: Figure 5 As shown in the flowchart in , the echo signal feature point detection includes detecting the time series point corresponding to the peak point of the echo signal function and retrieving the time series point corresponding to the peak point of the first-order derivative of the envelope function.

[0141] The time series point corresponding to the peak point detection of the envelope function is the feature point 1, which is expressed as:

[0142] Result[i1]suject to R p (i1)=max(R p (i))

[0143] Among them, Result[i1] represents the retrieval result of feature point 1.

[0144] The time series point corresponding to the peak point retrieval of the first-order derivative of the envelope function is the feature point 2, which is expressed as:

[0145]

[0146] Among them, Result[i2] represents the search result of feature point 1. Indicates the envelope R p (i) The first derivative of

[0147] Step 3: Construct three objective functions

[0148] The specific implementation is as follows: Figure 5 As shown in the flowchart in , the three objective functions are respectively minimizing the loss function value of the actual echo signal and the fitted exponential function signal, minimizing the time series point difference value of the feature point 1 of the actual echo signal and the fitted exponential function signal, and minimizing the time series point difference value of the feature point 2 of the actual echo signal and the fitted exponential function signal.

[0149] In objective function 1, the expression of the minimum loss function value of constructing the actual echo signal and the fitted exponential function signal is:

[0150]

[0151] Among them, Loss(x) is the loss function when the echo signal parameters are x = [A, m, T, τ], S f is the fitting function.

[0152] In objective function 2, the expression for minimizing the time series point difference between the actual echo signal and the fitted exponential function signal is:

[0153]

[0154] in, Retrieve the difference between the fitting function signal and the feature point 1 of the actual signal when the echo signal parameters are x = [A, m, T, τ], R f is the envelope function of the fitting function signal.

[0155] In objective function 3, the expression for minimizing the time series point difference between the actual echo signal and the fitted exponential function signal is:

[0156]

[0157] in, is the characteristic point 2 of the fitting function signal and the actual signal when the echo signal parameters are x = [A, m, T, τ]

[0158] Time Series

[0159] Point search difference.

[0160] The final optimization goal based on the three objective functions is:

[0161] minL(x)=[L(x)1,L(x)2,L(x)3]

[0162]

[0163] Among them, L(x)1, L(x)2, and L(x)3 are three objective functions.

[0164] Step 4: Obtain the optimal solution of the multi-objective function

[0165] The specific implementation is as follows: Figure 5 and Figure 6 As shown in the flowchart in , the key steps in using the OMOPSO method to obtain the optimal solution of the multi-objective function include particle velocity, archive size and crowding determination.

[0166] The speed of each example is expressed as:

[0167]

[0168] Among them, C1 and C2 of OMOPSO are not fixed values. Their calculation method is shown in formula (10). ω is randomly selected from [0.1, 0.5], c1 and c2 are randomly selected from [1.5, 2], r1 and r2 are random numbers uniformly distributed between 0 and 1, and C1 and C2 are control and gbes t The acceleration constant that affects the particle velocity. In addition, w t is the inertia weight of particle i and controls the trade-off between global history and local history. and is the position and velocity of particle i at time t.

[0169] After obtaining the velocity of each particle, the position of particle i is updated. The updated position of the particle i can be expressed as:

[0170]

[0171] The Archive size is expressed as:

[0172]

[0173] The size of the Archive is determined by ∈-dominance, where the ∈ value limits the Archive size and the value of ∈ is generally defined by the user. If ∈ satisfies the following formula, the decision vector x1 can be regarded as ∈-dominance x2.

[0174] The crowding density adopts NSGA-II and its expression method is as follows:

[0175]

[0176] Among them, i d represents the crowding degree of the i-th non-dominated particle in Archive, represents the jth objective function value at point i+1, represents the j-th objective function value at point i-1, and m is the number of objective functions (the value is 3).

[0177] Step 5: Calculate the contamination depth

[0178] The specific implementation is as follows: Figure 5 The calculation of the pollutant depth is shown in the flowchart in the ToF method, where ToF is the time delay in x = [A, m, T, τ], and the specific method is as follows:

[0179]

[0180] Where ToF1 and ToF2 are the transit times of ultrasound under dry road surface and under pollutants, respectively, h is the depth of pollutants, v u is the speed of ultrasound propagation in air.

Claims

1. A vehicle-mounted airport runway pollutant detection method based on ultrasonic technology, characterized in that: The method is based on the following detection system: Vehicle-mounted ultrasonic sensor; The vehicle-mounted ultrasonic sensor is installed on the vehicle, and includes an ultrasonic air probe, which adopts a single-transmitting and single-receiving air-coupled probe. The probe is embedded in a PCB circuit board, and the PCB board is connected to the rotating motor; The vehicle-mounted ultrasonic sensor is installed on the vehicle, including a single-transmitting and single-receiving air-coupled ultrasonic air probe. The sensor is equipped with a lifting and rotating motor to control the rotating probe to detect the height and angle. The lifting motor realizes the lifting of the probe through a screw rod. The probe is embedded in a PCB circuit board and connected to the rotating motor. A waterproof cover is arranged outside the probe and the motor for waterproof and dustproof, and an aviation plug is arranged outside the waterproof cover for signal transmission. The sensor is installed on the vehicle through a fixture consisting of a detachable steel connecting rod, a fixed metal plate and hand screws. The steel rod can be adjusted in length to suit different installation requirements.

2. The vehicle-mounted airport runway pollutant detection method based on ultrasonic technology according to claim 1 is characterized in that: The pollutant detection information includes pollutant type and depth information. The vehicle-mounted ultrasonic sensor first performs type identification and then performs depth calculation. The type identification includes obtaining echo signal samples of various pollutants, extracting echo signal waveform characteristic parameters, constructing a CNN deep learning model, adjusting CNN model parameters, and outputting pollutant type identification results.

3. The vehicle-mounted airport runway pollutant detection method based on ultrasonic technology according to claim 2 is characterized in that: The echo signal waveform parameters include the amplitude, distortion degree and signal-to-noise ratio of the echo signal.

4. The vehicle-mounted airport runway pollutant detection method based on ultrasonic technology according to claim 3 is characterized in that: The echo signal amplitude is expressed as: A=max[S p (i)] Among them, A represents the amplitude of the echo signal, S p (t) represents the maximum voltage amplitude in the obtained echo signal time series point, i is the time series point of the echo signal; The degree of distortion is represented by cosine similarity, which is expressed as: Among them, S p ,S d Represents the actual ultrasonic echo signal, S d Represents the echo signal of the standard mixed exponential model, and N represents the total number of signal sampling points The echo signal of the mixed exponential model is specifically: Among them, t i is a time series point, A0 is the amplitude of the signal, m,T is the intrinsic parameter of the ultrasonic probe, which is used to characterize the waveform skewness and kurtosis of the signal, τ is the offset of the starting point, f0 is the center frequency, and θ is the initial phase; The signal-to-noise ratio of the echo signal is expressed as: Among them, P s and P n represent the signal power and noise power respectively.

5. The vehicle-mounted airport runway pollutant detection method based on ultrasonic technology according to claim 2 is characterized in that: The CNN deep learning model includes a convolutional layer, an input layer, a hidden layer, a fully connected layer and an output layer. The convolutional layer is used to extract local features in the echo signal, and the fully connected layer is used to map the extracted features to the pollutant category. The neural network input is the echo signal waveform parameters, and the output is the type of pollutant.

6. The vehicle-mounted airport runway pollutant detection method based on ultrasonic technology according to claim 2 is characterized in that: The depth detection method includes echo signal envelope extraction, echo signal feature point retrieval, construction of a multi-objective function, obtaining the optimal solution of the multi-objective function using the OMOPSO method and pollutant depth calculation.

7. The vehicle-mounted airport runway pollutant detection method based on ultrasonic technology according to claim 6 is characterized in that: The echo signal envelope extraction process uses wavelet denoising, Hilbert transform and cubic spline curve smoothing methods to filter the echo signal to obtain a smooth signal, specifically: The process of wavelet denoising includes using Nakagami wavelet basis function, selecting wavelet coefficients, decomposition scale and threshold, using soft threshold method to remove signal noise, and finally reconstructing the signal; The Hilbert transform is: z(i)=H{S p (i)} R 0p (i)=|z(i)| Among them, H{S p (i)} is the Hilbert operation performed on the echo signal, z(i) is a complex signal, and |z(i)| is the modulus of the complex signal. The cubic spline interpolation method is: Among them, R p (i) is the interpolation curve, that is, the smoothed envelope, and s(i) is the cubic spline basis function. Among them, the basis function s(i) can be expressed as: s(i)=a i +b i (x-x i )+c i (x-x i )2+d i (x-x i ) Among them, the coefficient a i , b i , c i , d i It is obtained through interpolation conditions, curve continuity conditions, and continuity conditions of the first-order derivative and second-order derivative of the curve.

8. The vehicle-mounted airport runway pollutant detection method based on ultrasonic technology according to claim 7 is characterized in that: The echo signal feature points include: The first characteristic point is the peak point of the echo signal envelope function, which is expressed as: Result[i1]suject to R p (i1)=max(R p (i)) Among them, Result[i1] represents the search result of the first feature point in the time series; The second characteristic point is the peak point of the first-order derivative of the echo signal envelope function, which is expressed as: Among them, Result[i2] represents the search result of the second feature point in the time series. Indicates the envelope R p (i) The first derivative of 9. The vehicle-mounted airport runway pollutant detection method based on ultrasonic technology according to claim 8 is characterized in that: The multi-objective function includes: The first objective function is to minimize the loss function value between the actual echo signal and the fitted exponential function signal, and its expression is: Among them, loss(x) is the loss function when the echo signal parameters are x = [A, m, T, τ], S f is the fitting function; The second objective function is to minimize the difference between the first characteristic point of the actual echo signal and the fitted exponential function signal in the time series, and its expression is: in, Retrieve the difference between the time series point of the fitting function signal and the first characteristic point of the actual signal under the echo signal parameter x = [A, m, T, -], R f is the envelope function of the fitting function signal; The third objective function, wherein the difference between the retrieval results of the actual echo signal and the second characteristic point of the fitted exponential function signal in the time series is the smallest, is expressed as: in, Retrieving a difference value for a time series point of a second characteristic point of a fitting function signal and an actual signal under the echo signal parameter x=[A, m, T, τ]; The final optimization goal based on the three objective functions is: minL(x)=[l(x)1,L(x)2,L(x)3] Among them, L(x)1, L(x)2, and l(x)3 are three objective functions.

10. The vehicle-mounted airport runway pollutant detection method based on ultrasonic technology according to claim 9 is characterized in that: The steps of obtaining the optimal solution of the multi-objective function by using the OMOPSO method include: initializing particle position and velocity parameters, updating and mutating particle position and velocity, storing solutions of non-dominated particles, judging output conditions, and outputting a Pareto optimal solution set; Among them, the speed of each particle is expressed as: Among them, C1 and C2 of OMOPSO are control and gbest t The acceleration constant that affects the particle velocity, C1 and C2 are random numbers in [0,2], ω is a random number, r1 and r2 are used to increase the randomness of the search process and are random numbers in [0,1]; w t is the inertia weight of particle i and controls the trade-off between global history and local history; and is the position and velocity of particle i at time t; After obtaining the velocity of each particle, the position of particle i is updated, and the updated particle position is expressed as: The Archive is a solution set of non-dominated solutions of the multi-objective function. The size of Archive is determined by ∈, and ∈ satisfies the following formula: The size of the Archive is determined by ∈-dominance, where the value of ∈ limits the size of the Archive, the value of ∈ is defined by the user, and the decision vector x1 is ∈-dominance x2; The crowding degree is the density of the solution of the multi-objective function in the search space, which is expressed as follows: Among them, i d represents the crowding degree of the i-th non-dominated particle in Archive, represents the jth objective function value at point i+1, represents the j-th objective function value at point i-1, and m is the number of objective functions.

Images