Non-contact rapid assessment method for residual gas mixture status in diving cylinders

Abstract

The present application belongs to the technical field of gas ratio assessment in diving cylinders, and relates to a non-contact rapid assessment method for the residual gas mixing state in diving cylinders. The present invention transmits sound waves of continuously varying frequencies through an acoustic wave transmitter to penetrate the cylinder wall and the residual gas inside. A receiver collects the signal and calculates the time difference to obtain the original sound velocity value. At the same time, the surface temperature and pressure data of the cylinder are collected for physical correction of the sound velocity to eliminate the influence of environmental factors and obtain a standardized sound velocity value. The acoustic wave signal is then filtered, the amplitude attenuation spectrum and the phase shift spectrum are calculated, the minimum attenuation frequency and the maximum phase difference are extracted, and the trained inversion model is called in combination with the standardized sound velocity value to predict the gas grouping ratio. There is no need to exhaust air or remove sensors, which improves efficiency and safety, reduces costs, and ensures accurate assessment of the helium ratio in the helium-oxygen mixture cylinder, thereby protecting the safety of the diver.

Description

Technical Field

[0001] The present application belongs to the technical field of diving cylinder gas ratio assessment, and more specifically, relates to a non-contact rapid assessment method for the residual gas mixing state in a diving cylinder. Background Art

[0002] Refilling diving cylinders is a critical step in diving activities and is directly related to the safety of divers. Traditional refilling methods usually require emptying the residual gas in the cylinder first, or inserting a sensor into the cylinder for detection to determine the residual gas ratio. However, these methods have the following major problems:

[0003] The process of evacuating the gas is time-consuming, especially when dealing with a large number of gas cylinders, which will significantly reduce the filling efficiency. Although the method of inserting sensors avoids evacuating the gas, the operation is complicated and each detection requires additional time and steps. Secondly, during the process of evacuating the gas, if the operation is improper, it may cause gas leakage, posing a safety hazard. In addition, inserting sensors into the gas cylinder also increases operational risks, especially in high-pressure environments. Frequent evacuation operations and the use of sensors will increase the cost of filling.

[0004] The above problems are particularly prominent for helium-oxygen mixture cylinders. Helium is an expensive gas, and the accuracy of its ratio assessment directly affects the subsequent filling ratio accuracy. If the residual helium ratio is assessed incorrectly, the gas ratio after filling may not meet the requirements, which in turn affects the breathing safety of divers and may even cause serious consequences such as decompression sickness or oxygen poisoning.

[0005] In summary, during the refilling process of diving cylinders, how to quickly and accurately evaluate the proportion of residual gas inside the cylinders, especially the helium-oxygen ratio in helium-oxygen mixed cylinders, has become a technical problem that needs to be solved urgently. Summary of the Invention

[0006] The present invention provides a non-contact rapid assessment method for the residual gas mixing state of a diving cylinder, which aims to achieve rapid and accurate assessment of the proportion of residual gas in the cylinder.

[0007] A non-contact rapid assessment method for the residual gas mixture state of a diving cylinder comprises the following steps:

[0008] The acoustic transmitter emits sound waves of continuously changing frequencies. After the sound waves penetrate the diving cylinder wall and the residual gas inside, the sound wave signals are collected in real time by the acoustic receiver. The time difference between the sound wave emission and reception is recorded, and the original sound velocity value is calculated based on the fixed transmission distance. The surface temperature and pressure data of the cylinder are also collected synchronously.

[0009] Based on the sound wave signal collected by the receiver in real time, noise interference is filtered out, and the amplitude attenuation spectrum and phase shift spectrum of the sound wave signal are calculated based on the filtered clean sound wave signal; then, the minimum attenuation frequency and maximum phase difference are calculated based on the amplitude attenuation spectrum and phase shift spectrum;

[0010] Based on the cylinder surface temperature and cylinder pressure data, the original sound speed value is physically corrected to obtain a standardized sound speed value that is independent of the environment;

[0011] Based on the minimum attenuation frequency, maximum phase difference and normalized sound velocity value, the trained inversion model is called to obtain the gas grouping prediction value.

[0012] The present invention transmits sound waves of continuously varying frequencies through an acoustic wave transmitter to penetrate the cylinder wall and residual gas inside. A receiver collects signals and calculates the time difference to obtain an original sound velocity value. Simultaneously, surface temperature and pressure data of the cylinder are collected for physical correction of the sound velocity to eliminate the influence of environmental factors and obtain a standardized sound velocity value. The acoustic wave signal is then filtered, an amplitude attenuation spectrum and a phase shift spectrum are calculated, the minimum attenuation frequency and the maximum phase difference are extracted, and a trained inversion model is called in combination with the standardized sound velocity value to predict the gas grouping ratio. This method does not require exhausting air or removing sensors, thereby improving efficiency and safety, reducing costs, and ensuring accurate assessment of the helium ratio in the helium-oxygen mixed cylinder, thereby protecting the safety of divers.

[0013] Preferably, filtering out noise interference includes the following steps:

[0014] The original sound wave signal is filtered using an FIR filter, where the FIR filter coefficients are designed using a Kaiser window;

[0015] Apply a Hanning window to the filtered signal, and multiply the filtered signal by the Hanning window to obtain a windowed signal, that is, a clean sound wave signal after filtering.

[0016] Preferably, the FIR filter adopts a 128-order filter, and the corresponding Kaiser window shape parameter is 8.6.

[0017] Preferably, obtaining the minimum attenuation frequency comprises the following steps:

[0018] Perform fast Fourier transform on the received signal and the transmitted signal respectively, calculate the spectrum of the received signal and the transmitted signal, and then calculate the attenuation of each frequency point based on this;

[0019] In a predetermined frequency band, a continuous frequency band is found that satisfies the requirement that the attenuation is less than a predetermined threshold, and the lowest frequency of the continuous frequency band is extracted as the minimum attenuation frequency.

[0020] Preferably, obtaining the maximum phase difference comprises the following steps:

[0021] Locate the predetermined frequency point, extract the phase angle of the received signal and the transmitted signal at the predetermined frequency point, calculate the difference between the phase angles of the received signal and the transmitted signal, and perform phase unwrapping processing. Finally, calculate the absolute value of the phase difference after unwrapping as the maximum phase difference.

[0022] Preferably, the physical correction of the original sound speed value comprises the following steps:

[0023] Convert the measured temperature from Celsius to Kelvin, and establish a polynomial regression model based on the experimental data to calculate the temperature correction factor. Multiply the original sound velocity value by the temperature correction factor to obtain the temperature-corrected sound velocity.

[0024] A polynomial regression model is established based on experimental data to calculate the pressure correction factor. The temperature-corrected sound velocity is multiplied by the pressure correction factor to obtain the pressure-corrected sound velocity, i.e., the normalized sound velocity value; or,

[0025] Apply the physical compensation formula to obtain the normalized sound speed.

[0026] Preferably, the inversion model adopts a lightweight neural network model, wherein the lightweight neural network model includes an input layer, two hidden layers and an output layer;

[0027] The input layer includes three nodes corresponding to the three features of the input vector, the first hidden layer includes 8 nodes, using the ReLU activation function, the second hidden layer includes 5 nodes, using the ReLU activation function, and the output layer includes n nodes corresponding to the mole fraction of the gas in n;

[0028] The output of the lightweight neural network model satisfies the following constraints: the sum of the mole fractions of all gases is 1, and the mole fraction of each gas ranges from 0 to 100%.

[0029] Preferably, in each detection cycle, online error compensation is performed, including the following steps:

[0030] In each detection cycle, the current gas composition prediction value, temperature and pressure are obtained. Using the reference sound velocity as a benchmark, the reference sound velocity is adjusted according to the changes in temperature and pressure. The impact on the sound velocity is then calculated based on the specific heat ratio and mole fraction of each gas component. The theoretical sound velocity is calculated based on this impact and the adjusted reference sound velocity.

[0031] The difference between the theoretical sound speed and the standardized sound speed is calculated to determine whether the absolute value of the difference is greater than a preset threshold. If it exceeds the preset threshold, the Kalman filter is used to correct the inversion model parameters.

[0032] Preferably, the Kalman filter corrects the inversion model parameters including the following steps:

[0033] Define the Kalman filter parameters: use the weight matrix of the inversion model as the state variable, and the normalized sound speed and gas composition prediction values ​​as the observation data; the state transition matrix adopts a constant model, the change of the weight matrix is ​​described by process noise, and the observation model uses the difference between the detected sound speed and the theoretical sound speed as the observation signal, and the observation noise describes the observation error;

[0034] Kalman filter update: Use the weight matrix of the previous moment to estimate the weight matrix of the current moment, and update the current prediction error covariance based on the prediction error covariance of the previous moment and the process noise covariance;

[0035] Calculate the Kalman gain based on the current prediction error covariance and the observation noise covariance, adjust the weight matrix based on the calculated Kalman gain; and update the error covariance based on the Kalman gain;

[0036] The updated weight matrix is ​​reloaded into the inversion model for subsequent gas composition prediction.

[0037] The beneficial effects of the present invention include:

[0038] The present invention transmits sound waves of continuously varying frequencies through an acoustic wave transmitter to penetrate the cylinder wall and residual gas inside. A receiver collects signals and calculates the time difference to obtain an original sound velocity value. Simultaneously, surface temperature and pressure data of the cylinder are collected for physical correction of the sound velocity to eliminate the influence of environmental factors and obtain a standardized sound velocity value. The acoustic wave signal is then filtered, an amplitude attenuation spectrum and a phase shift spectrum are calculated, the minimum attenuation frequency and the maximum phase difference are extracted, and a trained inversion model is called in combination with the standardized sound velocity value to predict the gas grouping ratio. This method does not require exhausting air or removing sensors, thereby improving efficiency and safety, reducing costs, and ensuring accurate assessment of the helium ratio in the helium-oxygen mixed cylinder, thereby protecting the safety of divers. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the embodiments or descriptions of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0040] Figure 1 This is a flowchart of the overall steps provided by an embodiment of the present invention.

[0041] Figure 2 This is a flow chart of sound velocity correction provided by an embodiment of the present invention.

[0042] Figure 3This is a structural diagram of an exemplary inversion model provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0043] In order to make the technical problems, technical solutions and beneficial effects to be solved by this application more clearly understood, this application is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0044] See also Figure 1 As shown in FIG, a non-contact rapid assessment method for the residual gas mixture state of a diving cylinder includes the following steps:

[0045] S1. The acoustic transmitter emits sound waves of continuously varying frequencies. After the sound waves penetrate the diving cylinder wall and the residual gas inside, the sound wave signals are collected in real time by the acoustic receiver. The time difference between the sound wave transmission and reception is recorded, and the original sound velocity value is calculated based on the fixed transmission distance. The cylinder surface temperature and cylinder pressure data are also collected simultaneously.

[0046] In this embodiment, the acoustic wave transmitter and the acoustic wave receiver are symmetrically installed on the diving cylinder, and during data acquisition, a trigger signal is synchronized by the main controller to simultaneously start the sweep frequency signal generation of the ultrasonic transmitter and the ADC sampling clock of the temperature and pressure sensors. That is, the main controller sends a hardware trigger pulse through the GPIO pin, and the pulse is ultrasonically transmitted and received by the temperature and pressure sensors. After receiving the trigger signal, the ultrasonic transmitter immediately starts the generation of the linear sweep signal; the ADC sampling clocks of the temperature and pressure sensors are also started synchronously to ensure that the temperature and pressure data and the acoustic wave signal are collected simultaneously.

[0047] The specific link for acoustic wave signal collection is as follows:

[0048] First, the acoustic wave transmitter generates a sweep frequency signal with a frequency linearly increasing from 0.5MHz to 1.5MHz according to the trigger signal of the main controller. The duration of the sweep frequency signal is 1ms to ensure that it has sufficient time to propagate in the helium-oxygen mixture. The generated sweep frequency signal is amplified by a power amplifier to ensure that the acoustic wave signal has sufficient strength to penetrate the helium-oxygen mixture. The specific amplification speed is not limited here and can be adjusted according to actual conditions. The amplified signal drives the piezoelectric transducer to generate mechanical vibration, thereby converting electrical energy into acoustic energy.

[0049] The specific linearly increased frequency sweep parameters mentioned above are only exemplary. It should be noted that frequency sweeping is based on increasing the penetration probability, that is, dynamically changing the signal frequency within a certain frequency range (such as continuous scanning from low frequency to high frequency), and screening out the frequency most suitable for penetrating the target medium (bottle wall) by testing the transmission effect of signals of different frequencies; compared with single frequency or non-uniform frequency sweeping, this method can avoid feature omissions due to incomplete frequency coverage, and provide continuous and complete frequency dimension data for the subsequent calculation of amplitude attenuation spectrum and phase shift spectrum, ensuring that key features such as minimum attenuation frequency can be accurately extracted, thereby improving the discrimination of gas mixture ratios.

[0050] The piezoelectric transducer at the receiving end captures the acoustic wave signal transmitted from the transmitting end. Since the sound speed of the helium-oxygen mixed gas is different from that of pure air, the acoustic wave signal will be affected by the gas composition during the propagation process, resulting in changes in the frequency, amplitude and phase of the signal. Since the received acoustic wave signal is very weak, it needs to be amplified and filtered by a signal conditioning circuit to improve the signal-to-noise ratio. The signal conditioning circuit includes a preamplifier and a low-pass filter for amplifying weak signals and filtering out high-frequency noise. This is a conventional technical means in this field, so it will not be described in detail in the present invention.

[0051] The conditioned signal is input to a dedicated audio processing chip (such as ADAU1772) for 24-bit ADC sampling. The sampling rate is set to at least 3.2MHz to meet the Nyquist theorem and ensure that the high-frequency components of the signal can be accurately captured. The audio processing chip records the start time of the transmitted signal and the arrival time of the received signal. Combined with the fixed transmission distance (that is, the installation distance between the sound wave transmitter and the sound wave receiver, since the diameter of the gas cylinder is known, the distance between the sound wave transmitter and the sound wave receiver is also known), the original sound speed value is calculated. The specific expression is as follows: ; Where: Indicates time difference; Indicates the arrival time of the received signal; Indicates the starting time of transmitting the signal; Indicates a fixed transmission distance; Indicates the original sound speed value.

[0052] Finally, the main controller needs to encapsulate the original sound velocity value, collected temperature and pressure, and original sound wave sampling data into a data packet for subsequent analysis.

[0053] Finally, the main controller needs to encapsulate the original sound velocity value, collected temperature and pressure, and original sound wave sampling data into a data packet for subsequent analysis.

[0054] S2. Based on the real-time acoustic signal collected by the receiver, noise interference is filtered out, and the amplitude attenuation spectrum and phase shift spectrum of the acoustic signal are calculated based on the filtered clean acoustic signal; the minimum attenuation frequency and the maximum phase difference are calculated based on the amplitude attenuation spectrum and the phase shift spectrum;

[0055] In this embodiment, a possible implementation method is:

[0056] Receive the original sound wave signal, exemplarily with a signal length of 3200 sampling points, corresponding to a duration of 1ms, a sampling rate of 3.2MHz, or 3.2 million sampling points per second, and a quantization accuracy of 24 bits; synchronously receive a reference signal, i.e., the original swept frequency signal from the transmitter, whose frequency varies linearly between 0.5MHz and 1.5MHz;

[0057] In this embodiment, the data is denoised by bandpass filtering and windowing, as follows:

[0058] Bandpass filtering removes frequency components below 0.4MHz and above 1.6MHz in the signal to reduce the impact of noise. Specifically, a 128-order finite impulse response (FIR) filter is used. The filter coefficients are calculated using a Kaiser window with a window parameter of β=8.6. The passband frequency range of the filter is ; Indicates that the filter only retains signals within a predetermined frequency range and removes signals of other frequencies;

[0059] For each sampling point n (n ranges from 0 to 3199), calculate the filtered signal value: Where: represents the filtered signal; represents the filter coefficient, which is designed based on the Kaiser window;

[0060] The Hanning window is applied to reduce spectrum leakage, and the filtered signal values ​​obtained by alignment are windowed. The specific expression is as follows: Where: represents the Hanning window coefficient; N represents the signal length, which is 3200; and n represents the sample index. Where: Represents the windowed signal, based on which the filtered clean sound wave signal is obtained.

[0061] Perform 4096-point fast Fourier transform on the received signal and the transmitted signal respectively, calculate the spectrum of the received signal and the transmitted signal, and then calculate the attenuation of each frequency point based on this spectrum; the specific expression is as follows: ; ; Where: represents the spectrum of the received signal; represents the spectrum of the reference signal; represents the received signal after windowing; represents the reference signal; k represents the frequency index, ranging from 0 to 4095; Indicates the index of the current time sampling point; Represents the complex exponential factor, which is the weight used to convert the time domain signal into the frequency domain signal; Indicates frequency The attenuation under the condition of L is fixed transmission distance. , is the frequency resolution, calculated based on the sampling rate and the number of FFT points;

[0062] Scan the 0.4-1.6MHz frequency band and find the A continuous frequency band is output and the lowest frequency of the band is output.

[0063] Locate the predetermined frequency point 1MHz (index ), extract the phase angle between the received signal and the transmitted signal at a predetermined frequency point: ; Where: Indicates the phase angle of the received signal at 1MHz; Indicates the phase angle of the reference signal at 1MHz; and represent the imaginary and real parts of the spectrum respectively; Represents the received signal spectrum at frequency index 1280; represents the spectrum of the reference signal at frequency index 1280;

[0064] Calculate the difference in phase angle between the received signal and the transmitted signal, perform phase unwrapping, and finally calculate the absolute value of the phase difference after unwrapping as the maximum phase difference. The expression is as follows: Where: Indicates the maximum phase difference; Represents the unwrapping function, which is used to process phase jumps and make the phase difference continuous.

[0065] This embodiment also includes an exception handling mechanism, which is as follows:

[0066] When satisfied When , the current frame data is discarded and re-acquisition is triggered;

[0067] When there is no satisfaction dB / cm continuous band, set , is the lowest attenuation frequency.

[0068] In this embodiment, the combination of a 128-order FIR filter and a β=8.6 Kaiser window achieves the synergy of high frequency selectivity and strong noise suppression. The 128-order FIR filter ensures an extremely narrow transition band, accurately locks the target frequency region, and prevents the signal to be retained from being filtered out. The selection of the β=8.6 Kaiser window ensures that the sidelobe attenuation in the stopband reaches 80dB, avoiding the problem of the noise to be filtered out not being filtered out completely. At the same time, the linear phase characteristics of the FIR and the smooth transition of the Kaiser window ensure that the amplitude and phase information of the target signal are not distorted.

[0069] S3. Based on the cylinder surface temperature and cylinder pressure data, perform physical correction on the original sound velocity value to obtain a standardized sound velocity value that is independent of the environment;

[0070] See also Figure 2 As shown, as an implementation of this embodiment, the physical correction of the original sound speed value includes the following steps:

[0071] Convert measured temperature from Celsius to Kelvin ,Right now , represents the measured temperature; and a polynomial regression model is established based on the experimental data to calculate the temperature correction factor, and the original sound velocity value is multiplied by the temperature correction factor to obtain the temperature-corrected sound velocity; ;

[0072] in: is the sparsity in the polynomial regression model, determined by fitting experimental data. In this embodiment, a third-order polynomial regression model is used; represents the temperature correction factor; ;

[0073] Where: represents the speed of sound after temperature correction; Indicates the original sound speed value;

[0074] A polynomial regression model is established based on the experimental data, and the pressure correction factor is calculated. The temperature-corrected sound velocity is multiplied by the pressure correction factor to obtain the normalized sound velocity. The polynomial regression model is similar to the polynomial regression model for the temperature correction factor, except that in this embodiment, a single-order polynomial or second-order polynomial regression model is used for the pressure correction polynomial regression model. The expression for the pressure-corrected sound velocity is as follows: ;

[0075] Where: represents the pressure correction factor; It represents the speed of sound after pressure correction; Represents the normalized speed of sound.

[0076] As another implementation of this embodiment, the normalized sound speed is obtained by applying a physical compensation formula, which is specifically expressed as follows: ;

[0077] Where: Indicates the reference temperature, the temperature under standard conditions, set to 293K (20℃); Indicates reference pressure, the pressure under standard conditions, set to 1 atmosphere (atm); It represents the ambient temperature in Kelvin, which is measured in real time by the sensor; P represents the ambient pressure in kilopascals, which is measured in real time by the sensor.

[0078] S4. Based on the minimum attenuation frequency, maximum phase difference, and normalized sound velocity value, call the trained inversion model to obtain the gas grouping prediction value.

[0079] See also Figure 3 As shown, the inversion model adopts a lightweight neural network model, wherein the lightweight neural network model includes an input layer, two hidden layers and an output layer;

[0080] The input layer includes three nodes corresponding to the three features of the input vector The first hidden layer includes 8 nodes and uses the ReLU activation function. The second hidden layer includes 5 nodes and uses the ReLU activation function. The output layer includes n nodes corresponding to the mole fractions of the gases in n. For example, n is 2, and the mole fractions of helium and oxygen are output. This is only an exemplary technical solution and is not limited in the present invention.

[0081] The output of the lightweight neural network model satisfies the following constraints: the sum of the mole fractions of all gases is 1, and the mole fraction of each gas ranges from 0 to 100%.

[0082] In each round of detection cycle, online error compensation is performed, including the following steps:

[0083] In each detection cycle, the current gas composition prediction value, temperature and pressure are obtained. Using the reference sound velocity as a benchmark, the reference sound velocity is adjusted according to the changes in temperature and pressure. The impact on the sound velocity is then calculated based on the specific heat ratio and mole fraction of each gas component. The theoretical sound velocity is calculated based on this impact and the adjusted reference sound velocity.

[0084] ;

[0085] Where: represents the theoretical sound speed calculated based on the gas dynamics formula; Indicates the reference sound speed, which is the sound speed under standard environmental conditions (such as 20°C, which is 1 standard atmospheric pressure); T indicates the temperature currently measured by the sensor; P indicates the pressure currently measured by the sensor; Indicates the reference temperature; Indicates reference pressure; represents the specific heat ratio of the gas component in the ith position, which is the ratio of the specific heat capacity at constant pressure to the specific heat capacity at constant volume; represents the mole fraction of the gas component in the i-th position;

[0086] The difference between the theoretical sound speed and the standardized sound speed is calculated to determine whether the absolute value of the difference is greater than the preset threshold (2m / s). If it exceeds the preset threshold, the Kalman filter is used to correct the inversion model parameters.

[0087] The Kalman filter performs correction of inversion model parameters and comprises the following steps:

[0088] Define the Kalman filter parameters: The weight matrix of the inversion model (such as and ) as the state variable, normalized sound speed and gas composition predictions As observation data; the state transfer matrix adopts a constant model , represents process noise; represents the state estimate (weight matrix) at the kth iteration; the change of the weight matrix is ​​described by the process noise, where the observation model uses the difference between the detected sound speed and the theoretical sound speed as the observation signal, and the observation noise describes the observation error: , represents the observed value; represents the observation noise;

[0089] Kalman filter update: Use the weight matrix of the previous moment to estimate the weight matrix of the current moment: ,in represents the predicted state estimate; Represents the state estimate after the last update;

[0090] And update the current forecast error covariance based on the forecast error covariance of the previous moment and the process noise covariance : ,in represents the process noise covariance; Represents the error covariance matrix after the last update;

[0091] Calculate the Kalman gain based on the current prediction error covariance and the observation noise covariance: ;in represents the observation matrix; represents the observation noise covariance matrix; represents the transpose of the observation matrix;

[0092] The weight matrix is ​​adjusted based on the calculated Kalman gain: ;in represents the updated state estimate;

[0093] And update the error covariance based on the Kalman gain; ;in represents the identity matrix;

[0094] The updated weight matrix is ​​reloaded into the inversion model for subsequent gas composition prediction.

[0095] In this embodiment, only the normalized sound velocity, minimum attenuation frequency, and maximum phase difference are used as input features. These three features are orthogonal and complementary. The normalized sound velocity reflects the overall elastic properties of the gas, the attenuation frequency reflects the gas absorption properties, and the phase difference reflects the propagation properties. These three features describe the gas composition from different dimensions, resulting in low information redundancy. Furthermore, after considering these three features, this embodiment eliminates the need to consider cylinder parameters. The specific reasons are as follows: the cylinder diameter is already a known parameter when calculating the original sound velocity, so it does not need to be repeatedly input. The impact of acoustic wave penetration shielding has already been addressed through signal filtering, and the design of the swept frequency signal ensures that it can effectively penetrate the cylinder wall and fully interact with the gas inside. Secondly, during the experimental training process, the cylinder material and volume are fixed, and their impact on the sound wave is included in the systematic error of the model training phase and does not need to be input as a variable. Furthermore, in this embodiment, the absence of cylinder parameters enhances the model's anti-interference ability. Due to the inherent relationship between gas composition and acoustic characteristics, and the fact that cylinder parameters (such as material aging and surface corrosion) may change over time, the inclusion of these parameters will increase measurement error. By eliminating the influence of the cylinder wall through physical correction and signal processing, the model focuses on the gas composition itself, resulting in higher stability.

[0096] Secondly, through sound speed correction, the standardized sound speed is used as a reference condition, so that the input features are only related to the gas composition, further reducing the dependence on the cylinder parameters; finally, the Kalman filter is used to dynamically correct the model weights through the difference between the sound speed and the standardized sound speed. Even if there is the influence of unmodeled cylinder parameters (such as bottle wall thickness deviation), the prediction accuracy can be guaranteed through adaptive adjustment.

[0097] The above are only preferred embodiments of the present application and are not intended to limit the present application. Any modifications, equivalent replacements, and improvements made within the spirit and principles of the present application should be included in the scope of protection of the present application.

Claims

1. A non-contact rapid assessment method for the residual gas mixture state of a diving cylinder, characterized in that: The following steps are involved: The acoustic wave transmitter emits sound waves of continuously changing frequencies. After the sound waves penetrate the wall of the diving cylinder and the residual gas inside, the sound wave signals are collected in real time by the acoustic wave receiver, which records the time difference between the sound wave emission and reception. The original sound speed value is then calculated based on the fixed transmission distance. And simultaneously collect the cylinder surface temperature and cylinder pressure data; Based on the sound wave signal collected by the receiver in real time, noise interference is filtered out, and the amplitude attenuation spectrum and phase shift spectrum of the sound wave signal are calculated based on the filtered clean sound wave signal; then, the minimum attenuation frequency and maximum phase difference are calculated based on the amplitude attenuation spectrum and phase shift spectrum; The filtering out of noise interference comprises the following steps: The original sound wave signal is filtered using an FIR filter, where the FIR filter coefficients are designed using a Kaiser window; Apply a Hanning window to the filtered signal, and multiply the filtered signal by the Hanning window to obtain a windowed signal, that is, a clean sound wave signal after filtering; Obtaining the minimum attenuation frequency comprises the following steps: Perform fast Fourier transform on the received signal and the transmitted signal respectively, calculate the spectrum of the received signal and the transmitted signal, and then calculate the attenuation of each frequency point based on this; In a predetermined frequency band, a continuous frequency band is searched for which the attenuation is less than a predetermined threshold, and the lowest frequency of the continuous frequency band is extracted as the minimum attenuation frequency; Based on the cylinder surface temperature and cylinder pressure data, the original sound speed value is physically corrected to obtain a standardized sound speed value that is independent of the environment; Based on the minimum attenuation frequency, maximum phase difference and normalized sound velocity, the trained inversion model is called to obtain the gas grouping prediction value; The inversion model adopts a lightweight neural network model, wherein the lightweight neural network model includes an input layer, two hidden layers and an output layer; The input layer includes three nodes corresponding to the three features of the input vector, the first hidden layer includes 8 nodes, using the ReLU activation function, the second hidden layer includes 5 nodes, using the ReLU activation function, and the output layer includes n nodes corresponding to the mole fraction of the gas in n; The output of the lightweight neural network model satisfies the following constraints: the sum of the mole fractions of all gases is 1, and the mole fraction of each gas ranges from 0 to 100%; In each round of detection cycle, online error compensation is performed, including the following steps: In each detection cycle, the current gas composition prediction value, temperature and pressure are obtained. Using the reference sound velocity as a benchmark, the reference sound velocity is adjusted according to the changes in temperature and pressure. The impact on the sound velocity is then calculated based on the specific heat ratio and mole fraction of each gas component. The theoretical sound velocity is calculated based on this impact and the adjusted reference sound velocity. The difference between the theoretical sound speed and the standardized sound speed is calculated to determine whether the absolute value of the difference is greater than a preset threshold. If it exceeds the preset threshold, the Kalman filter is used to correct the inversion model parameters.

2. The non-contact rapid assessment method for the residual gas mixture state of a diving cylinder according to claim 1, characterized in that: The FIR filter adopts a 128-order filter, and the corresponding Kaiser window shape parameter is 8.

6.

3. The non-contact rapid assessment method for the residual gas mixture state of a diving cylinder according to claim 1, characterized in that: Obtaining the maximum phase difference comprises the following steps: Locate the predetermined frequency point, extract the phase angle of the received signal and the transmitted signal at the predetermined frequency point, calculate the difference between the phase angles of the received signal and the transmitted signal, and perform phase unwrapping processing. Finally, calculate the absolute value of the phase difference after unwrapping as the maximum phase difference.

4. The non-contact rapid assessment method for the residual gas mixture state of a diving cylinder according to claim 1, characterized in that: The physical correction of the original sound speed value comprises the following steps: Convert the measured temperature from Celsius to Kelvin, and establish a polynomial regression model based on the experimental data to calculate the temperature correction factor. Multiply the original sound velocity value by the temperature correction factor to obtain the temperature-corrected sound velocity. A polynomial regression model is established based on experimental data to calculate the pressure correction factor. The temperature-corrected sound velocity is multiplied by the pressure correction factor to obtain the pressure-corrected sound velocity, i.e., the normalized sound velocity value; or Apply the physical compensation formula to obtain the normalized sound speed.

5. The non-contact rapid assessment method for the residual gas mixture state of a diving cylinder according to claim 1, characterized in that: The Kalman filter performs correction of inversion model parameters and comprises the following steps: Define the Kalman filter parameters: use the weight matrix of the inversion model as the state variable, and the normalized sound speed and gas composition prediction values ​​as the observation data; the state transition matrix adopts a constant model, the change of the weight matrix is ​​described by process noise, and the observation model uses the difference between the detected sound speed and the theoretical sound speed as the observation signal, and the observation noise describes the observation error; Kalman filter update: Use the weight matrix of the previous moment to estimate the weight matrix of the current moment, and update the current prediction error covariance based on the prediction error covariance of the previous moment and the process noise covariance; Calculate the Kalman gain based on the current prediction error covariance and the observation noise covariance, adjust the weight matrix based on the calculated Kalman gain; and update the error covariance based on the Kalman gain; The updated weight matrix is ​​reloaded into the inversion model for subsequent gas composition prediction.

Images