An aircraft hydraulic oil identification method based on gas detection technology

Abstract

The application discloses an aircraft hydraulic oil identification method based on a gas detection technology, utilizes high-sensitivity organic mass spectrum detection technology, quickly and efficiently obtains fingerprints of various samples of an aircraft, and establishes a standard hydraulic oil spectrum library; a set of standardized aircraft hydraulic oil identification methods are formed, real-time mass spectrum is compared and analyzed with spectrum in the standard hydraulic oil spectrum library, quick detection is realized, and through combination of multiple advanced data processing methods, not only the timeliness of detection is improved, but also the environmental adaptability of the method is enhanced, and accurate identification and quantitative analysis of the hydraulic oil are ensured.

Description

Technical Field

[0001] This invention belongs to the technical field of oil and gas leak detection, specifically relating to a method for identifying aircraft hydraulic oil based on gas detection technology. Background Technology

[0002] The aircraft hydraulic system, as the core power source for flight control, landing gear retraction and extension, and braking systems, directly impacts flight safety in terms of its reliability. Hydraulic oil, as the energy transfer medium, must meet stringent performance requirements, including high wear resistance, oxidation resistance, and low-temperature fluidity. Different hydraulic oil specifications are used under different operating conditions. Mixing or using substandard hydraulic oil can lead to seal swelling and failure, corrosion of metal components, or deterioration of foaming properties, ultimately causing major malfunctions such as hydraulic system leaks and actuator jamming. Therefore, accurately identifying the type of hydraulic oil is a crucial step in ensuring the safe operation of the aircraft's hydraulic system during routine maintenance.

[0003] Currently, the industry's conventional detection technologies for hydraulic oil identification mainly include the following three categories:

[0004] 1. Physical property analysis method: This method involves measuring physical parameters such as viscosity, density, and flash point of the oil. It relies on laboratory-grade precision instruments (such as rotational viscometers and densitometers), and a single test can take over 30 minutes. Furthermore, it cannot distinguish between hydraulic oils with similar chemical additives and is prone to measurement inaccuracies due to oil oxidation and deterioration.

[0005] 2. Chemical Reagent Reaction Method: This method uses specific reagents to react with functional additives such as phosphate esters and corrosion inhibitors in hydraulic oil to produce a colorimetric reaction. While this method has the potential for rapid on-site detection, it suffers from drawbacks such as stringent reagent storage conditions (requiring protection from light and refrigeration), significant sensitivity to environmental temperature and humidity, and an inability to quantitatively analyze the oil mixing ratio.

[0006] 3. Spectroscopic analysis: This method uses Fourier transform infrared (FTIR) or Raman spectroscopy to analyze the vibrational characteristic peaks of oil molecules. While this technique can accurately identify the chemical composition of oil, the equipment is expensive, requires professional personnel to analyze the spectra, and oil particles can easily contaminate the optical window, making it difficult to use stably in field environments such as hangars.

[0007] The limitations of existing technologies are mainly reflected in the following aspects: insufficient timeliness: the laboratory testing process usually takes 2-4 hours, which cannot meet the timeliness requirements of "inspection upon arrival" in line maintenance.

[0008] Poor environmental adaptability: Precision instruments are sensitive to vibration, temperature and humidity, and are difficult to work stably in complex working conditions such as airport tarmac and outdoor fields;

[0009] Economic drawbacks: The high maintenance costs of high-precision testing equipment result in low adoption rates among small and medium-sized aviation enterprises;

[0010] Low level of intelligence: The lack of automated data comparison modules and reliance on manual experience for judgment can easily lead to misjudgment risks. Therefore, there is an urgent need to develop a hydraulic oil field identification technology that combines rapid response, high accuracy (>98%), strong environmental robustness, and controllable cost to improve aircraft maintenance efficiency and reduce systemic risks caused by oil misuse.

[0011] Therefore, in view of the problems of insufficient timeliness and poor environmental adaptability of the existing technology, the present invention discloses an aircraft hydraulic oil identification method based on gas detection technology. Summary of the Invention

[0012] This invention discloses a method for identifying aircraft hydraulic oil based on gas detection technology, in order to solve the problems of insufficient timeliness and poor environmental adaptability in the existing hydraulic oil identification and detection process.

[0013] This invention is achieved through the following technical solution:

[0014] A method for identifying aircraft hydraulic oil based on gas detection technology includes the following steps:

[0015] Step 1: Use a polytetrafluoroethylene probe to collect target gas and ambient gas, and then enrich the collected gas before passing it into a mass analyzer.

[0016] Step 2: Use the dynamic baseline correction algorithm to correct the mass data obtained by the mass analyzer to obtain the corrected target gas signal and ambient gas signal;

[0017] Step 3: Use wavelet transform background signal elimination algorithm to eliminate the influence of ambient gas signal on target gas signal to obtain target gas signal after interference removal, and then perform filtering and smoothing processing on target gas signal;

[0018] Step 4: Identify the characteristic peaks of the target gas signal using variational mode decomposition and Bayesian optimized Gaussian mixture model;

[0019] Step 5: Use a similarity measurement algorithm to match the identified feature peaks with the reference peaks in the standard material spectrum of the standard hydraulic oil spectrum library to identify the hydraulic oil.

[0020] To better realize the present invention, step 2 further includes:

[0021] Step 2.1: Use the sliding window method to analyze the naive data and calculate the mean of the naive data in each time period to detect the trend of baseline drift;

[0022] Step 2.2: Based on the baseline data obtained after dynamic adjustment of the sliding window length, a Butterworth filter is used to perform low-pass filtering on the data to obtain a denoised smooth baseline signal;

[0023] Step 2.3: Based on the denoised smoothed baseline signal, the baseline drift model is obtained by fitting with the least squares method or the daughter-in-law polynomial model;

[0024] Step 2.4: Based on the original naive data and baseline drift model, the baseline correction formula is used to perform baseline correction on each data point to obtain the baseline-corrected naive data.

[0025] To better realize the present invention, the baseline drift model is further defined as follows:

[0026] B(t) = a0 + a1t + a2t 2 ;

[0027] Where: B(t) represents the baseline drift signal after fitting; t represents time; a0, a1, and a2 represent the fitting parameters, respectively.

[0028] To better realize the present invention, the dynamic adjustment of the sliding window length further includes: adjusting the sliding window length based on the fluctuation period, or adjusting the sliding window length based on the noise level, or adjusting the sliding window length based on the rate of change, or adjusting the sliding window length based on a weighted sum of the fluctuation period, noise level, and rate of change.

[0029] To better realize the present invention, step 4 further includes:

[0030] Step 4.1: Use variational state decomposition to decompose the smoothed rudimentary data into several intrinsic mode functions (IMFs). Optimize the decomposition function to further decompose the IMFs into IMF components with different frequency components.

[0031] Step 4.2: For each IMF component, fit the shape of the characteristic peak using a Gaussian mixture model, and estimate the parameters of the Gaussian mixture model through Bayesian optimization, thereby obtaining the optimal Gaussian mixture model parameters for each IMF component.

[0032] Step 4.3: Based on the optimal Gaussian mixture model parameters and the time-frequency characteristics of the signal, construct the feature vector for the peak;

[0033] Step 4.4: Using a similarity measurement method combining dynamic time warping and Procrustes analysis, feature vectors are matched. The matching degree between each feature peak and the reference peak is obtained through similarity measurement. The final matching peak is selected according to the set threshold.

[0034] To better realize the present invention, the optimized decomposition function is further defined as follows:

[0035]

[0036] Where: u k The k-th eigenmode function represents the signal components within a certain frequency band; w k This represents the center frequency corresponding to the k-th IMF component; δ(t) represents the derivative with respect to time t; δ(t) represents the Dirac function; and δ represents the unit impulse signal. This represents the Hilbert transform kernel, used to construct analytic signals and extract instantaneous frequencies; This represents a complex modulation term.

[0037] To better realize the present invention, further, in step 4, the concentration of the target gas is quantified by peak area integration for the identified characteristic peaks, specifically as follows:

[0038] Step A1: Calculate the baseline signal for each characteristic peak, and dynamically adjust the integration range using the characteristic peak parameters obtained through Gaussian fitting;

[0039] Step A2: Based on the dynamically adjusted integration range and the parameters obtained by Gaussian fitting, the numerical integration method is used to divide the integration interval into multiple small intervals, calculate the area between each small interval, and then sum the calculated areas to obtain the area of ​​the final characteristic peak.

[0040] Step A3: Based on the area of ​​the obtained characteristic peak, establish a standard curve between the area of ​​the characteristic peak and the concentration of the target gas through linear regression, and derive the concentration value of the target gas based on the established standard curve.

[0041] To better realize the present invention, in step 1, a dual-channel polytetrafluoroethylene probe is used to simultaneously collect the target gas and the ambient gas. After the concentration of the target gas and the ambient gas is enriched by 10-100 times by a gas enrichment device, the target gas and the ambient gas are then introduced into a mass analyzer.

[0042] To better realize the present invention, the gas enrichment device further adopts a two-stage spiral gas enrichment tube, and the inner wall of the two-stage spiral gas enrichment tube is coated with a polyether ether ketone coating.

[0043] To better realize the present invention, the construction step of the standard hydraulic oil spectrum library in step 5 is further as follows: take n ml of hydraulic oil sample and place it in a headspace vial, and connect the headspace vial opening to the sample inlet of a proton transfer reaction mass spectrometer through a Luer connector and a polyether ether ketone transfer tube. Perform proton transfer reaction mass spectrometry analysis on the sampled sample to obtain the mass spectrometry analysis results, and store the mass spectrometry analysis results in the obtained standard hydraulic oil spectrum library.

[0044] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0045] This invention utilizes high-sensitivity organic mass spectrometry to rapidly and efficiently acquire fingerprint spectra of various aircraft samples, establishing a standard hydraulic oil spectrum library. It forms a standardized method for identifying aircraft hydraulic oil, achieving rapid detection by comparing real-time mass spectra with those in the standard hydraulic oil spectrum library. Furthermore, the combination of multiple advanced data processing methods not only improves the timeliness of detection but also enhances the method's environmental adaptability, ensuring accurate identification and quantitative analysis of hydraulic oil. Attached Figure Description

[0046] Figure 1 This is a schematic diagram of the process steps of the present invention;

[0047] Figure 2 This is a schematic diagram of the preliminary analysis experiment for characteristic spectral peaks;

[0048] Figure 3 This is a peak result diagram of hydraulic oil. Detailed Implementation

[0049] Example 1:

[0050] This embodiment provides a method for identifying aircraft hydraulic oil based on gas detection technology, such as... Figure 1 As shown, it includes the following steps:

[0051] Step 1: Use a polytetrafluoroethylene probe to collect target gas and ambient gas, and then enrich the collected gas before passing it into a mass analyzer.

[0052] Step 2: Use the dynamic baseline correction algorithm to correct the mass data obtained by the mass analyzer to obtain the corrected target gas signal and ambient gas signal;

[0053] Step 3: Use wavelet transform background signal elimination algorithm to eliminate the influence of ambient gas signal on target gas signal to obtain target gas signal after interference removal, and then perform filtering and smoothing processing on target gas signal;

[0054] Step 4: Identify the characteristic peaks of the target gas signal using variational mode decomposition and Bayesian optimized Gaussian mixture model;

[0055] Step 5: Use a similarity measurement algorithm to match the identified feature peaks with the reference peaks in the standard material spectrum of the standard hydraulic oil spectrum library to identify the hydraulic oil.

[0056] Furthermore, in step 1, a dual-channel polytetrafluoroethylene probe is used to simultaneously collect the target gas and the ambient gas. After the concentration of the target gas and the ambient gas is enriched by 10-100 times by a gas enrichment device, the target gas and the ambient gas are then introduced into a mass analyzer.

[0057] Furthermore, step 2 specifically includes:

[0058] Step 2.1: Use the sliding window method to analyze the naive data and calculate the mean of the naive data in each time period to detect the trend of baseline drift;

[0059] Step 2.2: Based on the baseline data obtained after dynamic adjustment of the sliding window length, a Butterworth filter is used to perform low-pass filtering on the data to obtain a denoised smooth baseline signal;

[0060] Step 2.3: Based on the denoised smoothed baseline signal, the baseline drift model is obtained by fitting with the least squares method or the daughter-in-law polynomial model;

[0061] Step 2.4: Based on the original naive data and baseline drift model, the baseline correction formula is used to perform baseline correction on each data point to obtain the baseline-corrected naive data.

[0062] Furthermore, step 4 specifically includes:

[0063] Step 4.1: Use variational state decomposition to decompose the smoothed rudimentary data into several intrinsic mode functions (IMFs). Optimize the decomposition function to further decompose the IMFs into IMF components with different frequency components.

[0064] Step 4.2: For each IMF component, fit the shape of the characteristic peak using a Gaussian mixture model, and estimate the parameters of the Gaussian mixture model through Bayesian optimization, thereby obtaining the optimal Gaussian mixture model parameters for each IMF component.

[0065] Step 4.3: Based on the optimal Gaussian mixture model parameters and the time-frequency characteristics of the signal, construct the feature vector for the peak;

[0066] Step 4.4: Using a similarity measurement method combining dynamic time warping and Procrustes analysis, feature vectors are matched. The matching degree between each feature peak and the reference peak is obtained through similarity measurement. The final matching peak is selected according to the set threshold.

[0067] Furthermore, in step 4, the concentration of the target gas is quantified by integrating the peak area of ​​the identified characteristic peaks, specifically as follows:

[0068] Step A1: Calculate the baseline signal for each characteristic peak, and dynamically adjust the integration range using the characteristic peak parameters obtained through Gaussian fitting;

[0069] Step A2: Based on the dynamically adjusted integration range and the parameters obtained by Gaussian fitting, the numerical integration method is used to divide the integration interval into multiple small intervals, calculate the area between each small interval, and then sum the calculated areas to obtain the area of ​​the final characteristic peak.

[0070] Step A3: Based on the area of ​​the obtained characteristic peak, establish a standard curve between the area of ​​the characteristic peak and the concentration of the target gas through linear regression, and derive the concentration value of the target gas based on the established standard curve.

[0071] Furthermore, the construction steps of the standard hydraulic oil spectrum library in step 5 are as follows: take n ml of hydraulic oil sample and place it in a headspace vial, and connect the headspace vial opening to the sample inlet of a proton transfer reaction mass spectrometer through a Luer connector and a polyether ether ketone transfer tube. Perform proton transfer reaction mass spectrometry analysis on the sampled sample to obtain the mass spectrometry analysis results, and store the mass spectrometry analysis results in the obtained standard hydraulic oil spectrum library.

[0072] Example 2:

[0073] This embodiment discloses a method for identifying aircraft hydraulic oil based on gas detection technology, which is a further optimization based on Embodiment 1, specifically as follows:

[0074] Data acquisition and preprocessing: A polytetrafluoroethylene probe with heating function is used to collect target gas and ambient gas respectively. The gas enters the mass spectrometer after passing through a pre-enrichment device. A dynamic baseline correction algorithm is used to correct the acquired data to obtain the corrected target gas signal and ambient gas signal.

[0075] Background signal elimination: Based on the collected ambient gas mass spectrometry data, wavelet transform background signal elimination technology is applied to subtract the influence of background signal from the target gas signal to obtain the target gas signal data after interference removal. It should be noted that in this embodiment, dual channels are used to acquire the target gas signal and ambient gas signal respectively. The target gas signal contains background signal. On this basis, the background signal is eliminated by wavelet transform background signal elimination technology.

[0076] Data filtering and smoothing: The target gas signal data after removing interference is smoothed using a Savitzky-Golay filter to obtain smoothed mass spectrometry data;

[0077] Peak identification and feature peak matching: Based on smoothed mass spectrometry data, feature peaks are identified by combining variational mode decomposition and Bayesian optimized Gaussian mixture model. Then, feature peak matching is achieved by using a similarity measurement method based on the identified feature peaks.

[0078] Peak area integral: For the identified characteristic peaks, the concentration of the target substance is quantified by integrating the peak area to obtain the peak area integral value;

[0079] Spectral similarity analysis: Determine the similarity between the collected spectrum and the spectrum of the target substance in the standard hydraulic oil spectrum library, and obtain the similarity value and judgment result.

[0080] This invention utilizes high-sensitivity organic mass spectrometry to rapidly and efficiently acquire fingerprint spectra of various aircraft samples, establishing a standard hydraulic oil spectrum library. A standardized method for identifying aircraft hydraulic oil is formed, achieving rapid detection by comparing real-time mass spectra with those in the standard hydraulic oil spectrum library. Furthermore, the combination of multiple advanced data processing methods not only improves the timeliness of detection but also enhances the method's environmental adaptability, ensuring accurate identification and quantitative analysis of hydraulic oil.

[0081] As one possible implementation of this embodiment, the dynamic baseline correction includes the following steps:

[0082] Dynamic baseline drift monitoring: The sliding window method is used to perform short-term analysis of mass spectrometry data. The trend of baseline drift is detected by calculating the signal mean in each time period. If the mean changes significantly, it indicates that the baseline has drifted. Based on this, the baseline data after the sliding window is obtained. The length of the sliding window is set according to the fluctuation frequency of the data.

[0083]

[0084] In the formula: Window Mean i The value represents the average signal value within the window; w represents the window length; x represents the average signal value within the window. j This represents the j-th data point in the mass spectrometry data sequence;

[0085] There are several ways to dynamically adjust the length of a sliding window, such as window adjustment based on fluctuation period, window adjustment based on noise level, window adjustment based on rate of change, and adjustment based on a weighted sum of the above three methods; the adjustment method can be selected according to the actual situation.

[0086] The window adjustment based on the fluctuation cycle is as follows:

[0087] w = k × T wave ;

[0088] In the formula: w represents the length of the sliding window; k represents a scaling constant; its value ranges from 2 to 10; it is used to control the width of the window. A smaller k value will result in a faster window response, while a larger k value will result in a smoother window; T wave The fluctuation period of the data is represented by the Fourier transform or periodic analysis of the signal.

[0089] Window adjustment based on noise level:

[0090]

[0091] In the formula: w noise σ represents the length of the sliding window after noise adjustment. data The standard deviation of data is used to measure the volatility of the data; σ noise The standard deviation of noise reflects the noise level in the data and is obtained by calculating the standard deviation after low-pass filtering the data.

[0092] Window adjustment based on rate of change:

[0093]

[0094] In the formula: w rate The sliding window length is indicated and adjusted based on the rate of change; Rate of Change represents the rate of change of the signal, indicating the degree of change of the signal, which is obtained by calculating the derivative of the signal or using a moving average filter.

[0095] By calculating the mean within the window, the direction of baseline drift can be dynamically determined. If the mean changes significantly, it indicates that the baseline has drifted and needs to be corrected.

[0096] Low-pass filtering denoising: Based on the baseline data after the sliding window, a Butterworth filter is used for low-pass filtering to obtain a smoothed baseline signal after denoising; by selecting the cutoff frequency and order of the Butterworth filter, most of the high-frequency noise is removed to obtain a smoothed baseline signal.

[0097] Baseline drift model construction: Based on the denoised smoothed baseline signal, the baseline drift signal is fitted using the least squares method with a linear or quadratic polynomial model to obtain the baseline drift model, as shown below:

[0098] B(t) = a0 + a1t + a2t 2 ;

[0099] In the formula: B(t) represents the baseline drift signal after fitting; t represents time; a0, a1, and a2 all represent fitting parameters;

[0100] Dynamic baseline correction: Based on the original mass spectrometry data and the baseline drift model, baseline correction is performed on each data point using the baseline correction formula to obtain the baseline-corrected mass spectrometry data, as shown below:

[0101] y ′ (t)=y(t)-B(t);

[0102] In the formula: y(t) represents the signal strength in the original data; y ′ (t) represents the corrected signal strength.

[0103] As one possible implementation of this embodiment, the specific steps of peak identification and feature peak matching are as follows:

[0104] Overlapping peak separation: The smoothed mass spectrum signal is decomposed into k intrinsic mode functions (LMFs) using variational mode decomposition (VMD). Each LMF represents a frequency component of the signal. The goal of VMD is to automatically decompose the signal into IMFs with different frequency components through an optimization process, which is achieved through the following optimization problem:

[0105]

[0106] In the formula: u k The k-th eigenmode function represents the signal components within a certain frequency band; w k This represents the center frequency corresponding to the k-th IMF; δ(t) represents the derivative with respect to time t; δ(t) represents the Dirac function; and δ represents the unit impulse signal. This represents the Hilbert transform kernel, used to construct analytic signals and extract instantaneous frequencies; Represents a complex modulation term;

[0107] Based on the above optimization objective of minimizing the difference between the smoothed mass spectrum signal and each intrinsic mode function, the following constraints are constructed:

[0108] ∑ k u k (t) = s(t);

[0109] In the formula: s(t) represents the smoothed mass spectrum signal;

[0110] Based on the above constraints, we ensure that the sum of all intrinsic magic tower functions obtained from the decomposition is recovered and the smoothed mass spectrum signal is restored.

[0111] The value of k is selected through spectral kurtosis analysis to determine the appropriate number of decompositions K. Specifically, we prioritize modal components with a frequency range of 0.5Hz to 2Hz to better separate overlapping peaks in the signal. The dynamic determination of the value of K is the total number of modes in the VMD decomposition. Its selection logic is based on the statistical characteristics of the signal spectral kurtosis, aiming to separate the true feature peaks in a way that minimizes redundancy.

[0112] For example, the original mass spectrum signal has two overlapping peaks (peak spacing 0.3 Da) near m / z = 194.

[0113] Calculating the SK curve revealed two significant high kurtosis frequency bands (corresponding to peak frequencies of 1.2 Hz and 1.5 Hz); initially, k was 3.

[0114] The three IMF components are decomposed into two true peaks: IMF1 and IMF2; IMF3 is noise (σ is too large and A is too small); after removing IMF3, the actual effective K = 2.

[0115] Multi-peak joint modeling: For each IMF component, a Gaussian mixture model (GMM) is used to fit the shape of the peak, mathematically represented as follows:

[0116]

[0117] In the formula: A k μ represents the weight of each Gaussian component. k Indicates the position of the peak; σ k represents the peak width; x represents the mass-to-charge ratio of the mass spectrum, which is the independent variable; n represents the number of Gaussian components, i.e., the hypothesized number of peaks;

[0118] In the Gaussian mixture model, a Bayesian optimization framework is used for parameter estimation, as follows:

[0119] A loss function is constructed to measure the fit between GMM and IMF, as shown in the following expression:

[0120] L=α·SSIM(G,IMF)+β·KL div (IMF,G);

[0121] In the formula: SSIM is the structural similarity index, which measures the structural similarity between GMM and IMF. The larger the SSIM, the better the fit of GMM; KL represents the KL divergence, which calculates the difference in statistical distribution between GMM and IMF. The smaller the KL divergence, the more similar the statistical characteristics of GMM are to the distribution of IMF; α and β represent weighting coefficients.

[0122] Setting a reasonable empirical distribution for GMM parameters based on domain knowledge:

[0123] For the peak position μ, a uniform distribution μ ~ U (peak position ± 0.2 Da) is set, that is, the search is performed within ± 0.2 Da of the expected position of the peak;

[0124] For the peak width σ, we define the Gamma distribution as σ ~ Gamma(2, 0.5); the Gamma distribution is suitable for representing the standard deviation of a normal distribution.

[0125] Initially, Bayesian optimization evaluates the loss function by randomly selecting several parameter sets and models the objective function using a Gaussian process;

[0126] As the optimization process proceeds, Bayesian optimization gradually adjusts the parameter combination based on the current model predictions and the results of the data collection function, explores an efficient parameter space, and ultimately finds the optimal GMM parameters.

[0127] The optimal Gaussian mixture model parameters for each IMF are obtained through Bayesian optimization, including peak position, peak width, and weight of each Gaussian component.

[0128] Dynamic feature fusion matching:

[0129] Based on the parameters obtained from CMM and the time-frequency characteristics of the signal, a 12-dimensional feature vector is constructed, including but not limited to: main peak position offset, peak width ratio, kurtosis and skewness, wavelet packet energy entropy, and Hilbert marginal spectrum peak value.

[0130] Peak matching is performed using a similarity metric that combines Dynamic Time Warping (DTW) with Procrustes analysis, as detailed below:

[0131] By combining DTW and Procrustes analysis, a comprehensive similarity metric is obtained, and the contributions of the two methods are balanced by setting a dynamically adjusted parameter λ.

[0132] Similarity = (1-λ)·DTW Score +λ·Procrustes Score ;

[0133] In the formula: DTW Score Indicates the score calculated by DTW; Procrustes Score This represents the score calculated by Procrustes analysis;

[0134] When the signal-to-noise ratio is low, below 10dB, λ = 0.7, DTW should be given priority.

[0135] When the signal-to-noise ratio is high, such as a signal-to-noise ratio greater than or equal to 20 dB, set λ = 0.3 and prioritize the Procrustes analysis results;

[0136] Based on the calculated Similarity and the set threshold, matching peaks are selected.

[0137] In this embodiment, by combining DTW and Procrustes analysis as similarity metrics, the accuracy of peak matching can be effectively improved when dealing with overlapping peaks in mass spectrometry data. Specifically, DTW is used to align signal timing, while Procrustes analysis is used to compare signal shapes, thereby avoiding matching errors caused by time misalignment or peak shape differences. By dynamically adjusting the weight parameter λ, this method can adaptively adjust the matching strategy under different noise environments, further improving the reliability and accuracy of peak matching.

[0138] Furthermore, in this embodiment, specifically in hydraulic oil detection, especially when facing adjacent peak interference, this scheme can effectively suppress interference and improve the signal-to-noise ratio, with a suppression ratio improvement of over 15dB. For example, in the detection of the main peak m / z=194 in aircraft hydraulic oil, when there is interference at m / z=195: two independent peaks (194.1±0.03 and 195.2±0.05) are separated by VMD, and the 194.08 peak is accurately identified through Bayesian optimization, with a confidence level of 0.92; the feature fusion similarity reaches 0.87, exceeding the preset threshold of 0.85, significantly improving the reliability of oil identification.

[0139] As one possible implementation of this embodiment, the specific steps of the peak area integration are as follows:

[0140] Based on the peak shape and noise level, an adaptive integration threshold is dynamically set. First, the baseline signal of each peak is calculated, and the standard deviation and peak height of the target peak are determined based on the Gaussian fitting results. Then, the integration threshold is calculated using the following formula:

[0141] T threshold =A×(1-α);

[0142] In the formula: α represents the adjustment parameter used to control the sensitivity of the integral threshold, 0 < α < 1; T threshold This represents the dynamically calculated integration threshold; A represents the maximum height of the target peak.

[0143] Based on the calculated threshold, the specific range of the peak area integral is determined. The integration range extends to the left and right sides of the characteristic peak until the signal intensity falls below the adaptive threshold. For a Gaussian distributed peak, the integration range is:

[0144] t left =μ-3σ,t right =μ+3σ;

[0145] In the formula: t left Indicates the left boundary of the peak; tright σ represents the right boundary of the peak; μ represents the center of the peak; 3σ represents the extension range.

[0146] After determining the integration range, the discrete points within the integration range are divided into smaller intervals. The trapezoidal rule is used to apply the formula to each smaller interval for integration, obtaining the contribution between each smaller interval. Finally, the contributions between all smaller intervals are summed to obtain the peak area A. peak :

[0147]

[0148] In the formula: G(t) i ) and G(t i+1 ) represent time points t respectively i and t i+1 Gaussian function value on; This represents the width between each cell; n represents the total number of time points.

[0149] Based on the obtained peak area, a standard curve between peak area and concentration is established through linear regression. The concentration value is then derived from the established standard curve, as shown in the following expression:

[0150] C = k × A peak +b;

[0151] In the formula: C represents the concentration of the target substance; k represents the slope of the standard curve; b represents the intercept of the standard curve.

[0152] In traditional peak area integration methods, the integration range of the peak is usually fixed. However, by using adaptive thresholding techniques, the integration range of the peak can be dynamically adjusted to cope with signal changes in different samples, thus enhancing robustness to background noise. Combining Gaussian fitting models with numerical integration methods not only improves the peak identification accuracy but also ensures the accuracy of peak area integration.

[0153] In this embodiment, peak identification and characteristic peak matching serve as the first screening step to help us identify potential target substances; peak area integration provides concentration data of the target substances for quantitative analysis, supplementing the qualitative judgment of peak matching; spectral similarity analysis provides further confirmation for qualitative analysis, ensuring that the target substances have a sufficiently high degree of matching with the standard spectral library in the entire spectrum.

[0154] As one possible implementation of this embodiment, in the data acquisition and preprocessing, the target gas and the environmental background are collected simultaneously through a dual-channel sampling system, wherein the gas enters the enrichment device at a concentration of 10-100 times and then enters the proton transfer reaction mass spectrometer.

[0155] The steps for constructing the standard hydraulic oil spectrum library are as follows:

[0156] Preliminary analysis of characteristic peaks in hydraulic oil:

[0157] Preliminary peak analysis experiments used headspace sampling to analyze the peaks formed in a proton transfer reaction mass spectrometer on hydraulic oil samples, such as... Figure 2 As shown. In the experiment, 10 ml of sample was placed in a headspace vial, and the headspace vial opening was connected to the mass spectrometer injection end via a Luer connector and a polyetheretherketone transfer tube (outer diameter 1 / 16 inch, length 0.5 m). In the experiment, the instrument sampling flow rate was set to 7.20 × 10 mL / d, the instrument integration time was 1 s, the ion source gas pressure was 502 Pa, the ion source current was 1 mA, and the transfer quadrupole RF voltage was set to 150 V.

[0158] Hydraulic oil samples were directly sampled and analyzed by proton transfer reaction spectroscopy. The peak results are as follows: Figure 3 As shown, the hydraulic oil sample exhibits numerous characteristic peaks, primarily distributed between m / z = 50 and m / z = 220, with the peak at m / z = 194 demonstrating high sensitivity and stability. This characteristic m / z value is absent in air, indicating that mass spectrometry analysis of the hydraulic oil sample yields corresponding high-sensitivity peaks, and the peak distribution differs significantly from that of air, possessing the specificity required for identification. These experiments demonstrate the feasibility of using proton transfer reaction mass spectrometry (PTMS) for identifying aircraft hydraulic oil.

[0159] Based on the above, a standard hydraulic oil spectrum library was established: characteristic mass spectra of hydraulic oil under different working conditions were obtained through laboratory calibration, and m / z = 194 ± 0.5 was determined as the main characteristic peak, with auxiliary peaks including m / z = 106, 120, 220, etc.

[0160] The gas pre-enrichment device employs a two-stage spiral gas enrichment tube with a polyether ether ketone coating on the inner wall. The tube is 1.2m long and 8mm in outer diameter. The gas is concentrated 10-100 times by the enrichment device before entering the proton transfer reaction mass spectrometer.

[0161] In summary, a breakthrough in aircraft hydraulic oil detection technology has been achieved by constructing a specific identification system based on the characteristic peak of m / z=194, combined with a multi-stage gas enrichment sampling device and an intelligent data analysis system. This technology improves the hydraulic oil identification accuracy to 98.5%, reduces the false alarm rate to 1.8% in complex industrial environments with multiple interference sources, achieves detection sensitivity at the 0.1 ppm level, and shortens the response time to less than 1 minute, successfully overcoming the contradiction between sensitivity and anti-interference performance in traditional proton transfer reaction mass spectrometry. Through an innovatively designed spiral gas enrichment device and dual-channel dynamic baseline correction technology, the target gas capture efficiency is increased to 92.5%, and the environmental interference elimination rate reaches 97%.

[0162] The other parts of this embodiment are the same as those in Embodiment 1, so they will not be described again.

[0163] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Any simple modifications or equivalent changes made to the above embodiments based on the technical essence of the present invention shall fall within the protection scope of the present invention.

Claims

1. A method for identifying aircraft hydraulic oil based on gas detection technology, characterized in that, Includes the following steps: Step 1: Use a polytetrafluoroethylene probe to collect target gas and ambient gas, and then enrich the collected gas before passing it into a mass spectrometer. Step 2: Use the dynamic baseline correction algorithm to correct the mass spectrometry data obtained by the mass spectrometer to obtain the corrected target gas signal and ambient gas signal; Step 3: Use wavelet transform background signal elimination algorithm to eliminate the influence of ambient gas signal from target gas signal to obtain target gas signal after interference removal, and then perform filtering and smoothing processing on target gas signal; Step 4: Identify the characteristic peaks of the target gas signal using variational mode decomposition and Bayesian optimized Gaussian mixture model; Step 4 specifically includes: Step 4.1: Use variational mode decomposition to decompose the smoothed mass spectrometry data into several intrinsic mode functions, and optimize the decomposition function to decompose the intrinsic mode functions into IMF components with different frequency components. Step 4.2: For each IMF component, fit the shape of the characteristic peak using a Gaussian mixture model, and estimate the parameters of the Gaussian mixture model through Bayesian optimization, thereby obtaining the optimal Gaussian mixture model parameters for each IMF component. Step 4.3: Based on the optimal Gaussian mixture model parameters and the time-frequency characteristics of the signal, construct the feature vector for the peak; Step 5: Use a similarity measurement algorithm to match the identified feature peaks with the reference peaks in the standard material spectrum of the standard hydraulic oil spectrum library to identify the hydraulic oil; use a similarity measurement method combining dynamic time warping and Procrustes analysis to match feature vectors, obtain the matching degree between each feature peak and the reference peak through similarity measurement, and filter out the final matching peaks according to the set threshold.

2. The aircraft hydraulic oil identification method based on gas detection technology according to claim 1, characterized in that, Step 2 specifically includes: Step 2.1: Analyze the mass spectrometry data using the sliding window method and calculate the mean mass spectrometry data for each time period to detect the trend of baseline drift; Step 2.2: Based on the baseline data obtained after dynamic adjustment of the sliding window length, a Butterworth filter is used to perform low-pass filtering on the data to obtain a denoised smooth baseline signal; Step 2.3: Based on the denoised smoothed baseline signal, the baseline drift model is obtained by fitting with the least squares method or a quadratic polynomial model. Step 2.4: Based on the original mass spectrometry data and baseline drift model, the baseline correction formula is used to perform baseline correction on each data point to obtain the baseline-corrected mass spectrometry data.

3. The aircraft hydraulic oil identification method based on gas detection technology according to claim 2, characterized in that, The baseline drift model is specifically as follows: ； Where: B(t) represents the baseline drift signal after fitting; t represents time; a0, a1, and a2 represent the fitting parameters, respectively.

4. The aircraft hydraulic oil identification method based on gas detection technology according to claim 3, characterized in that, The dynamic adjustment of the sliding window length includes: adjusting the sliding window length based on the fluctuation period, or based on the noise level, or based on the rate of change, or adjusting the sliding window length based on a weighted sum of the fluctuation period, noise level, and rate of change.

5. The aircraft hydraulic oil identification method based on gas detection technology according to claim 1, characterized in that, The optimized decomposition function is specifically: ； in: Let represent the k-th eigenmode function, which represents the components of the signal in a certain frequency band; This represents the center frequency corresponding to the k-th IMF component; This represents the derivative with respect to time t; represents the Dirac function, and represents the unit impulse signal; This represents the Hilbert transform kernel, used to construct analytic signals and extract instantaneous frequencies; This represents a complex modulation term.

6. The aircraft hydraulic oil identification method based on gas detection technology according to claim 5, characterized in that, In step 4, the concentration of the target gas is quantified by integrating the peak area of ​​the identified characteristic peaks, specifically as follows: Step A1: Calculate the baseline signal for each characteristic peak, and dynamically adjust the integration range using the characteristic peak parameters obtained through Gaussian fitting; Step A2: Based on the dynamically adjusted integration range and the parameters obtained by Gaussian fitting, the numerical integration method is used to divide the integration interval into multiple small intervals, calculate the area between each small interval, and then sum the calculated areas to obtain the area of ​​the final characteristic peak. Step A3: Based on the area of ​​the obtained characteristic peak, establish a standard curve between the area of ​​the characteristic peak and the concentration of the target gas through linear regression, and derive the concentration value of the target gas based on the established standard curve.

7. A method for identifying aircraft hydraulic oil based on gas detection technology according to any one of claims 1-4, characterized in that, In step 1, a dual-channel polytetrafluoroethylene probe is used to simultaneously collect the target gas and the ambient gas. After the concentration of the target gas and the ambient gas is enriched by 10-100 times by a gas enrichment device, the target gas and the ambient gas are then introduced into a mass spectrometer.

8. The aircraft hydraulic oil identification method based on gas detection technology according to claim 7, characterized in that, The gas enrichment device employs a two-stage spiral gas enrichment tube, the inner wall of which is coated with a polyether ether ketone coating.

9. A method for identifying aircraft hydraulic oil based on gas detection technology according to any one of claims 1-4, characterized in that, The construction steps of the standard hydraulic oil spectrum library in step 5 are as follows: take n ml of hydraulic oil sample and place it in a headspace vial, and connect the headspace vial opening to the sample inlet of a proton transfer reaction mass spectrometer through a Luer connector and a polyether ether ketone transfer tube. Perform proton transfer reaction mass spectrometry analysis on the sampled sample, obtain the mass spectrometry analysis results, and store the mass spectrometry analysis results in the obtained standard hydraulic oil spectrum library.

Images