Valve volatile organic compound leakage laser detection system and detection method

Abstract

The invention relates to the technical field of industrial safety monitoring and nondestructive testing, and discloses a valve volatile organic compound leakage laser detection system and method, and the system comprises a main detection TDLAS module which carries out the optical scanning of a target region, and obtains a concentration signal; the laser-induced micro-perturbation module synchronously applies physical micro-perturbation to excite an acoustic event; the multi-channel acoustic sensing array module captures acoustic signals, the precision synchronous control and data acquisition module uniformly controls and acquires acousto-optic multi-mode data, the central signal processing and intelligent decision-making unit processes the data, and intelligent decision-making and accurate positioning are realized through a feature extraction and deep learning model. By adopting the technical scheme of actively applying physical perturbation to excite leakage characteristics and synchronously acquiring and intelligently fusing response signals of an acousto-optic mode, the problems of low signal-to-noise ratio, easiness in interference, incapability of accurately positioning and dependence on manual interpretation in a complex industrial environment in the prior art are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of industrial safety monitoring and nondestructive testing, in particular to a valve volatile organic compound leakage laser detection system and method. BACKGROUND

[0002] In the industrial fields of petrochemical, natural gas processing and fine chemical, valves are the most widely used pipeline control elements. However, due to frequent operation, aging or corrosion of the seal, etc., valves have become one of the main sources of volatile organic compounds (VOCs) leakage. These leaks not only cause material loss and environmental pollution, but also pose a huge potential threat to production safety due to their flammable and explosive characteristics. Therefore, developing a technology system that can quickly and accurately detect valve area leakage is crucial for ensuring industrial production safety and fulfilling environmental protection responsibilities.

[0003] Currently, there are various detection technologies for such leaks, but they often face their own challenges in practical applications. For example, the technology based on tunable diode laser absorption spectroscopy (TDLAS) has high selectivity and sensitivity for specific gas molecules, but its conventional open optical path configuration can only give an integral value of the gas concentration along the optical path. This means that when the alarm is triggered, the staff knows that there is a leak along the optical path, but it is difficult to quickly locate the specific position of the leak on the valve body, flange or valve stem. The subsequent investigation work is still heavy. On the other hand, acoustic detection technology detects by picking up the ultrasonic waves generated by high-speed gas leakage. The problem with this method is that industrial sites are usually filled with strong background noise generated by pumps, motors and other equipment, and the acoustic signals emitted by weak leaks are easily overwhelmed by these noises. This results in the technology either failing to report due to too low signal-to-noise ratio or producing false alarms due to the inability to effectively distinguish between leakage sound and background sound in complex industrial environments.

[0004] In summary, the existing technology still has obvious limitations in the face of the core problems of small leaks, strong noise interference and precise positioning of the leakage source. SUMMARY

[0005] In view of the deficiencies of the prior art, the present application provides a valve volatile organic compound leakage laser detection system and method, which solves the problem of high sensitivity, high reliability detection and high precision positioning of volatile organic compound weak leakage generated by valves and other equipment in complex industrial noise environments.

[0006] To achieve the above purpose, the present application is realized by the following technical solutions:

[0007] The first aspect of the present application provides a valve volatile organic compound leakage laser detection system, comprising:

[0008] a main detection TDLAS module for generating a modulated detection laser beam to scan a target region of a valve and receiving the detection laser beam after passing through a gas region to be detected to obtain a second harmonic signal related to a volatile organic compound concentration;

[0009] a laser-induced micro-perturbation module for synchronously applying controllable physical micro-perturbations to a scanning point or a precise position adjacent to the scanning point when the detection laser beam reaches the scanning point;

[0010] a multi-channel acoustic sensing array module for synchronously capturing acoustic signals directly generated by a valve leakage or induced by the physical micro-perturbations;

[0011] a precise synchronization control and high-speed data acquisition module for precisely controlling scanning of the main detection TDLAS module, synchronously triggering the laser-induced micro-perturbation module, and synchronously acquiring the second harmonic signal and the acoustic signal to form time-series multi-modal data;

[0012] a central signal processing and intelligent decision unit for processing the time-series multi-modal data, extracting acousto-optic features, and fusing the acousto-optic features for intelligent decision, confirmation, and positioning of a leakage event.

[0013] Preferably, the main detection TDLAS module comprises:

[0014] a tunable diode laser for generating the detection laser beam;

[0015] a two-dimensional scanning galvanometer system for driving the detection laser beam to scan the target region of the valve;

[0016] a photodetector for receiving the detection laser beam after passing through the gas region to be detected;

[0017] a phase-locked amplifier for demodulating a signal received by the photodetector to extract the second harmonic signal related to the volatile organic compound concentration.

[0018] Preferably, the laser-induced micro-perturbation module comprises:

[0019] an auxiliary pulsed laser for generating laser pulses according to a trigger signal;

[0020] a micro-gas nozzle for jetting micro-gas flow pulses;

[0021] a high-speed precise solenoid valve for controlling the micro-gas nozzle to jet micro-gas flow pulses according to a trigger signal;

[0022] wherein the laser pulses and the micro-gas flow pulses are used as the controllable physical micro-perturbations;

[0023] a perturbation energy guiding unit for precisely guiding the laser pulse or the micro-puff pulse to a target point.

[0024] Preferably, the multi-channel acoustic sensing array module comprises:

[0025] at least three microphones for synchronously picking up the acoustic signal;

[0026] a plurality of low-noise preamplifiers respectively connected to the microphones for amplifying the picked-up acoustic signal;

[0027] a plurality of anti-aliasing filters respectively connected to the low-noise preamplifiers for filtering the amplified acoustic signal.

[0028] Preferably, the central signal processing and intelligent decision unit comprises:

[0029] a feature extraction unit for extracting the acousto-optic features from the time-series multi-modal data, wherein the acousto-optic features comprise TDLAS features related to the volatile organic compound concentration, time-domain and frequency-domain features of the acoustic signal, acoustic array features, and acousto-optic synchronism and correlation features, and a fusion feature vector is constructed based on the acousto-optic features;

[0030] a deep learning model inference unit for receiving the fusion feature vector and outputting a leakage state probability of each scanning point through a pre-trained deep neural network model;

[0031] a leakage confirmation and positioning unit for confirming a leakage event and refining a leakage source position based on the leakage state probability and preset acousto-optic time synchronism, acousto-optic shape correlation, and acousto-optic spatial consistency criteria.

[0032] The second aspect of the present application provides a valve volatile organic compound leakage laser detection method, applied to the valve volatile organic compound leakage laser detection system, comprising the following steps:

[0033] S1, scanning and synchronous perturbation data acquisition step: controlling the detection laser beam of the main detection TDLAS module to scan each point of the valve target area, synchronously triggering the laser-induced micro-perturbation module to apply controllable physical micro-perturbation when the detection laser beam reaches each scanning point, and synchronously collecting the second harmonic signal related to the volatile organic compound concentration and the acoustic signal captured by the multi-channel acoustic sensing array module by the precise synchronous control and high-speed data acquisition module, forming time-series multi-modal data of each scanning point;

[0034] S2, after obtaining the time-series multi-modal data of the scanning points, signal preprocessing and feature extraction are performed on the time-series multi-modal data: the time-series multi-modal data are preprocessed, and TDLAS features, acoustic features, and sound-light synchronization and correlation features are extracted, thereby constructing a fusion feature vector for each scanning point;

[0035] S3, after constructing the fusion feature vector for each scanning point, the fusion feature vector is used for intelligent leakage decision: the fusion feature vector is input into a pre-trained deep learning model, and a leakage state probability of each scanning point is output;

[0036] S4, after obtaining the leakage state probability of each scanning point, a leakage event confirmation and positioning step is performed according to the leakage state probability: a leakage event is confirmed in combination with a preset criterion, and a confirmed leakage source is refined and preliminarily quantified.

[0037] Preferably, in the S1 step, the synchronous acquisition of the second harmonic signal related to the volatile organic matter concentration and the acoustic signal captured by the multi-channel acoustic sensing array module includes: synchronously recording a time sequence of the second harmonic signal and a multi-channel original time sequence of the acoustic signal within a preset time window before, during and after the physical micro-disturbance event is triggered, and adding accurate time stamps to all collected data streams.

[0038] Preferably, in the S2 step, constructing a fusion feature vector for each scanning point includes: extracting the TDLAS features, the acoustic features, and the sound-light synchronization and correlation features, wherein:

[0039] The TDLAS features include a pre-disturbance steady-state concentration indicator, a disturbance-induced signal dynamic change feature, and a frequency domain feature;

[0040] The acoustic features include time domain features, frequency domain features, and mel-frequency cepstral coefficients of a single-channel acoustic signal, and arrival time difference and sound source positioning parameters based on a multi-channel acoustic signal;

[0041] The sound-light synchronization and correlation features include a time delay and a shape cross-correlation coefficient between TDLAS response features and acoustic event features.

[0042] Preferably, in the S3 step, the step of inputting the fusion feature vector into a pre-trained deep learning model and outputting the leakage state probability includes:

[0043] Preliminary encoding of features of different modalities is performed by using a parallel multi-modal feature encoding module inside the deep learning model;

[0044] The encoded multi-modal features are fused through an attention mechanism or a splicing operation inside the deep learning model;

[0045] A classification output layer of the deep learning model is used to output a posterior probability distribution of each scanning point belonging to a predefined leakage state category;

[0046] The leakage state probability is the posterior probability distribution.

[0047] Preferably, after the S4 step is performed, the method further comprises:

[0048] S5, a system adaptive adjustment and optimization step: according to the leakage state probability obtained in the current round of detection and the confirmed leakage event, dynamically adjusting the perturbation parameters of the laser-induced micro-perturbation module and the scanning path or scanning point density of the main detection TDLAS module;

[0049] S6, after the system adaptive adjustment and optimization step is completed, the steps S1 to S5 are repeatedly executed, or the leakage diagnosis result is finally output when the preset detection condition is met.

[0050] The present application provides a valve volatile organic compound leakage laser detection system and detection method. It has the following beneficial effects:

[0051] 1、The present application introduces an active acousto-optic synchronous detection mechanism, which uses controllable physical micro-perturbation to excite a potential leakage point, and then synchronously captures the response signals of two modalities of light and sound. This design achieves the technical effect of high sensitivity and high reliability in identifying weak leakage signals. Compared with the existing technology which only relies on single optical absorption or passive acoustic monitoring, it solves the problem of low signal-to-noise ratio and easy interference of background noise in complex industrial environments, which leads to frequent false negatives or false alarms.

[0052] 2、The present application has precise leakage source positioning capability. The system accurately controls the application position of micro-perturbation through a two-dimensional scanning galvanometer, and uses the time difference information of a multi-channel acoustic perception array to spatially resolve the sound source excited by the perturbation, achieving high-precision spatial positioning of the leakage source. Compared with traditional open optical path laser detection technology, it overcomes the inherent defect that it can only provide line integral concentration and cannot give the precise position of the leakage point, and also improves the insufficient positioning accuracy of traditional passive acoustic positioning.

[0053] 3、The application realizes the improvement in the intelligence and automation of diagnosis, adopts a deep learning model to deeply fuse sound and light multi-modal features, and combines physical rules to confirm the probability output by the model, thereby realizing intelligent decision of the leakage event, compared with the scheme of outputting original data or simple threshold alarm in the prior art, the application solves the problems of high dependence on manual secondary data interpretation, complicated diagnosis process and easy misjudgment, and makes the whole detection process more efficient and reliable. BRIEF DESCRIPTION OF DRAWINGS

[0054] Figure 1 is a flow framework diagram of the system of the application;

[0055] Figure 2 is a framework diagram of the main detection TDLAS module of the application;

[0056] Figure 3 is a framework diagram of the laser-induced micro-disturbance module of the application;

[0057] Figure 4 is a framework diagram of the multi-channel acoustic perception array module of the application;

[0058] Figure 5 is a flow framework diagram of the central signal processing and intelligent decision unit of the application;

[0059] Figure 6 is a flow diagram of the method of the application. DETAILED DESCRIPTION

[0060] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the drawings in the specification of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the protection scope of the application.

[0061] Please refer to the drawings in the specification of the application Figure 1 The embodiment of the application provides a valve volatile organic compound leakage laser detection system, which comprises:

[0062] Please refer to the drawings in the specification of the application Figure 2 The main detection TDLAS module is used for generating a modulated detection laser beam to scan the valve target area, and receiving the detection laser beam after passing through the gas area to be detected to obtain a second harmonic signal related to the volatile organic compound concentration.

[0063] The core function of the main detection TDLAS module is to generate a detection laser beam modulated in a specific way, and to accurately scan the valve target area where leakage may occur by using the detection laser beam.

[0064] During the scanning process, the detection laser beam passes through the area of ​​the gas to be measured around the valve. The main detection TDLAS module then receives and processes the laser beam after it has passed through this area, thereby acquiring a second harmonic signal that characterizes the concentration of volatile organic compounds.

[0065] In one specific implementation, the main detection TDLAS module physically includes a tunable diode laser, a two-dimensional scanning galvanometer system, a photodetector, and a lock-in amplifier. These components are connected via electrical signals and optical paths, working together to achieve the aforementioned functions.

[0066] A tunable diode laser is the source for detecting the laser beam. It is precisely controlled by an external laser driver controller. This controller provides the laser with a composite drive current, which is typically composed of a low-frequency scanning signal and a high-frequency modulation signal superimposed on each other.

[0067] For example, the low-frequency scanning signal can be a triangular wave or a sawtooth wave signal, which drives the center output wavelength of the laser to periodically scan within a small range to fully cover a specific absorption line of the target volatile organic molecule.

[0068] Meanwhile, the high-frequency modulation signal is typically a sinusoidal signal, which modulates the instantaneous output wavelength of the laser at high frequency. This modulation is the foundation for wavelength modulation spectroscopy (WMS) and is also key to the subsequent extraction of the second harmonic signal.

[0069] The two-dimensional scanning galvanometer system includes an X-axis scanning galvanometer and a Y-axis scanning galvanometer. It receives instructions from the system's precision synchronization control and high-speed data acquisition module to drive the detection laser beam generated by the tunable diode laser to perform two-dimensional point-by-point scanning of the valve target area according to a preset path.

[0070] The photodetector can be an indium gallium arsenide (InGaAs) detector with high bandwidth and high sensitivity. Its photosensitive surface is placed in the optical path after the detection laser beam passes through the region of the gas to be measured. Its function is to convert the received optical signal, whose intensity has been attenuated by gas absorption, into a corresponding electrical signal in real time.

[0071] The lock-in amplifier (LPA) is the core of the signal processing. Its electrical signal input is connected to the output of the photodetector. Simultaneously, its reference signal input receives a high-frequency sinusoidal signal of the same frequency and phase as the signal used to modulate the driving laser. Utilizing the phase-sensitive detection principle, the LPA can accurately demodulate and extract the amplitude of the second harmonic (2f) component from the noisy signal output by the photodetector. This amplitude is the second harmonic signal in this invention.

[0072] The underlying physical principle of this process is based on Beer-Lambert's law, which states that the intensity of transmitted light after passing through an absorbing medium can be expressed as:

[0073] I(ν)=I0(ν)exp[-PS(T)φ(ν)CL] (1);

[0074] Where I(ν) is the transmitted light intensity, I0(ν) is the incident light intensity, P is the total gas pressure, S(T) is the spectral line intensity at temperature T, φ(ν) is the line type function, C is the concentration of the gas to be measured, L is the effective absorption path length, and exp[-PS(T)φ(ν)CL] is the transmittance.

[0075] After high-frequency sinusoidal modulation, the instantaneous frequency ν(t) of the laser can be expressed as:

[0076]

[0077] in, It is the center frequency controlled by the low-frequency scanning signal, a is the modulation amplitude, f is the modulation frequency, π is the mathematical constant pi, and t is time.

[0078] When the gas concentration is low, the exponential term in equation (1) can be Taylor expanded. The derivation shows that the amplitude X of the second harmonic component in the signal received by the photodetector... 2f It is directly proportional to the concentration C of the gas being measured: X 2f ∝I0S(T)LC(3), where I0 is the incident light intensity. Therefore, the amplitude of the second harmonic signal output by the lock-in amplifier can directly and sensitively reflect the integrated concentration of volatile organic compounds on the scanning path. This second harmonic signal is then transmitted to the precision synchronous control and high-speed data acquisition module for digitization, serving as the key optical signal input for subsequent multimodal data fusion analysis. Through the above structure and method, the main detection TDLAS module can effectively convert gas concentration information into a stable and quantifiable electrical signal, providing a reliable data foundation for the high-precision leak detection of the entire system.

[0079] Please see the appendix Figure 3 The laser-induced micro-perturbation module is used to synchronously apply controllable physical micro-perturbations to the scanning point or its precise neighboring position when the detection laser beam arrives at each scanning point.

[0080] When the detection laser beam of the main detection TDLAS module reaches a certain scanning point in the target area of ​​the valve, the laser-induced micro-perturbation module can apply a physical micro-perturbation with controllable energy and time to the scanning point or its preset nearby precise position in strict synchronization.

[0081] In one specific embodiment, the laser-induced micro-perturbation module physically and integrally comprises an auxiliary pulsed laser, a gas jetting unit consisting of a micro-gas nozzle and a high-speed precision solenoid valve, and a common perturbation energy guiding unit. These components work in coordination under the unified scheduling of a precision synchronous control and high-speed data acquisition module.

[0082] The auxiliary pulsed laser is used to generate laser pulses with high peak power and short pulse width according to the trigger signal from the precision synchronous control and high-speed data acquisition module.

[0083] Exemplarily, a Q-switched Nd:YAG laser can be used, which generates laser pulses capable of releasing energy within nanoseconds.

[0084] When the laser pulse is directed to the target point and absorbed by the volatile organic gas molecules to be detected, the gas will experience a sharp local thermal expansion, thereby generating an acoustic pressure wave centered at the point. This process is the photoacoustic effect, and the generated acoustic pressure wave constitutes one form of controllable physical micro-perturbation.

[0085] The micro-gas nozzle has its outlet connected to an inert gas source through a high-speed precision solenoid valve. The opening and closing of the solenoid valve are also controlled by the trigger signal from the precision synchronous control and high-speed data acquisition module. When the solenoid valve is triggered to open momentarily, a pulse of micro-gas flow will be ejected at high speed through the nozzle.

[0086] This pulse of micro-gas flow exerts a momentary change in momentum and pressure on the target point, constituting another form of controllable physical micro-perturbation. This perturbation can interact with the weak gas flow that may exist at the valve leakage, thereby exciting or enhancing the acoustic characteristics of the leakage, making it more easily captured by the multi-channel acoustic sensing array module.

[0087] In one embodiment of the present application, both the laser pulse and the pulse of micro-gas flow are optional sources of physical micro-perturbation. The system can selectively trigger one of them or trigger both in a specific time sequence according to the detection strategy or environmental conditions, to achieve the best excitation effect on different types or states of leakage sources.

[0088] The perturbation energy guiding unit is used to accurately direct the energy generated by the selected perturbation source to the target point. When the laser pulse is selected as the perturbation source, the unit can include an optical focusing system consisting of a set of lenses and mirrors, used to focus the laser pulse and achieve precise spatial alignment with the detection laser beam of the main detection TDLAS module.

[0089] When the micro-gas flow pulse is used as the disturbance source, the unit mainly embodies the precise mechanical positioning and orientation of the micro-gas nozzle to ensure that the gas flow pulse can accurately act on the target scanning point.

[0090] In the whole system operation process, the working logic of the module is closely coupled with the scanning process of the main detection TDLAS module. First, the system determines the position of the current scanning point. Then, while the main detection TDLAS module measures the point, the precise synchronous control and high-speed data acquisition module sends a synchronous trigger signal. Finally, the laser-induced micro-disturbance module executes the signal to apply physical micro-disturbance.

[0091] Through this active and synchronous micro-disturbance mechanism, the application can create an acoustic event with a clear causal relationship and time marker at each measurement point. This not only enhances the detectability of weak leakage signals, but also provides a data basis for subsequent central signal processing and intelligent decision unit to distinguish real leakage signals from background noise using the acoustic-optical synchronicity and correlation features.

[0092] Please refer to the attached Figure 4 , a multi-channel acoustic sensing array module for synchronously capturing acoustic signals directly generated by valve leakage or induced by physical micro-disturbance;

[0093] In order to realize effective capture and accurate positioning of weak leakage signals, the application further discloses a specific embodiment of a multi-channel acoustic sensing array module. The multi-channel acoustic sensing array module in the application has the core function of synchronously capturing acoustic signals generated in the target area of the valve. The sources of these acoustic signals can be dual: both stable or non-stable noise directly generated by continuous leakage of the valve, and acoustic events induced by the laser-induced micro-disturbance module actively applying physical micro-disturbance.

[0094] In a specific embodiment, the multi-channel acoustic sensing array module includes at least three microphones, a plurality of low-noise preamplifiers corresponding to each of the microphones, and a plurality of anti-aliasing filters connected to the preamplifiers.

[0095] The at least three microphones are arranged in a pre-set geometric array with known coordinates near the target area of the valve. Exemplarily, such a configuration can be a triangular array in a two-dimensional plane or a tetrahedral array in a three-dimensional space. The use of an array of more than two microphones is the technical basis for spatial positioning of the acoustic signal source.

[0096] The output of each microphone is connected to the input of a low-noise preamplifier. The role of the amplifier is to linearly amplify the original acoustic electrical signal picked up by the microphone, which is usually very weak in energy, to boost the amplitude of the signal to the level range required by the subsequent data acquisition device, while introducing as little additional noise as possible.

[0097] The output of each low-noise preamplifier is further connected to the input of an anti-aliasing filter. The filter is a low-pass filter, which plays a role in filtering out invalid high-frequency components with a frequency higher than the Nyquist frequency before the signal is digitized and sampled, so as to avoid signal aliasing distortion in subsequent digital signal processing.

[0098] After amplification and filtering, the multiple parallel analog acoustic signals are transmitted to a precision synchronous control and high-speed data acquisition module. The module performs strict synchronous high-speed analog-to-digital conversion on the multiple signals to ensure that the acoustic data of all channels are accurately aligned in time, which is a prerequisite for subsequent sound source positioning algorithm calculation.

[0099] The physical model of sound source positioning is based on the principle of Time Difference of Arrival (TDOA). Assuming that the position of the sound source is S (x, y, z), and the position of the i-th microphone is known M i (x i ,y i ,z i ). The distance from the sound source to the i-th microphone is d i .

[0100]

[0101] If the j-th microphone is taken as the reference, the time difference τ ij between the arrival of the sound wave at the i-th microphone and the j-th microphone can be expressed as:

[0102]

[0103] where τ ij is the time difference that can be obtained by cross-correlation calculation of the acoustic signals of different channels; t i and t j are the times when the sound wave arrives at the i-th and j-th microphones, respectively; c s is the propagation speed of the sound wave in the air medium, which is a known or measurable environmental parameter.

[0104] At least two independent time differences (e.g., τ 12 and τ 13), a hyperbolic equation set about the unknown sound source position (x, y, z) can be constructed. By solving the equation set, the spatial position of the sound source can be calculated. The calculation process is performed in the central signal processing and intelligent decision unit.

[0105] In summary, the multi-channel acoustic sensing array module can provide multi-channel, high signal-to-noise ratio, time-synchronized acoustic signal data streams through its array structure and signal conditioning circuit. This not only realizes sensitive capture of acoustic events, but also provides key array feature information for precise positioning of the leakage source through the TDOA principle.

[0106] The precise synchronization control and high-speed data acquisition module is used for accurately controlling the scanning of the main detection TDLAS module, synchronously triggering the laser-induced micro-perturbation module, and synchronously collecting the second harmonic signal and the acoustic signal to form time-series multi-modal data.

[0107] The precise synchronization control and high-speed data acquisition module disclosed in the application has the core function of unified time axis management of the entire detection process. It accurately controls the scanning action of the main detection TDLAS module, strictly synchronously triggers the laser-induced micro-perturbation module, and synchronously collects the second harmonic signal and the multi-channel acoustic signal, and finally forms structured time-series multi-modal data.

[0108] In a specific embodiment, the module can have a field programmable gate array (FPGA) or a high-performance microcontroller (MCU) as its core processing unit. The core processing unit is integrated with or externally connected to multiple digital-to-analog conversion (DAC) channels and multiple synchronous analog-to-digital conversion (ADC) channels, and is connected to the central signal processing and intelligent decision unit through a high-speed data interface.

[0109] In terms of control function, the module first generates a series of two-dimensional coordinate points according to the preset scanning path planning. For each coordinate point, the module outputs accurate control voltage signals through its digital-to-analog conversion channels to drive the X-axis and Y-axis galvanometer of the two-dimensional scanning galvanometer system, respectively, so as to guide the detection laser beam to the accurate position of the coordinate point.

[0110] In the synchronous triggering function, when the detection laser beam stably reaches the target scanning point, the module outputs an accurate digital trigger pulse with nanosecond-level jitter from a separate digital I / O port. This pulse signal is sent to the laser-induced micro-perturbation module to trigger the application of physical micro-perturbation.

[0111] The time of the digital trigger pulse is the time zero of the current data acquisition. The module uses its built-in multi-channel synchronous ADC driven by the same master clock to perform strict synchronous analog-to-digital conversion on the second harmonic signal from the main detection TDLAS module and the multi-channel acoustic signal from the multi-channel acoustic sensing array module.

[0112] To ensure the absolute comparability of different modal data in time, the module attaches a high-precision timestamp to each data sample or each data frame collected. The timestamp is derived from a high-frequency counter inside the module, which provides a unified system time reference during the entire system operation.

[0113] Finally, the module integrates the multi-channel data stream collected for each scanning point into a structured data packet, which is the time-series multi-modal data. For the i-th scanning point, its corresponding time-series multi-modal data can be defined as:

[0114]

[0115] Where:

[0116] T start is the high-precision timestamp that identifies the start of this collection;

[0117] S 2f (t) is the digitized time series of the second harmonic signal recorded within the collection time window;

[0118] S mic1 (t) is the digitized time series of the acoustic signal recorded by the k-th microphone channel within the same collection time window;

[0119] N is the total number of microphones in the multi-channel acoustic sensing array module.

[0120] In the above manner, the module converts a complex detection behavior involving multiple physical processes into a series of standardized multi-modal data packets containing rich time-series and spatial information. These data packets are then transmitted in real time to the central signal processing and intelligent decision unit through a high-speed interface, providing a high-quality raw data basis for subsequent feature extraction, data fusion, and intelligent decision-making.

[0121] Please refer to the attached Figure 5 , the central signal processing and intelligent decision unit, for processing time-series multi-modal data, extracting acoustic and optical features, and fusing acoustic and optical features for intelligent decision-making, confirmation, and positioning of leakage events.

[0122] The central signal processing and intelligent decision unit is the data processing and decision center of the detection system of the application, and all its functions can be realized on a high-performance computer, an embedded processing platform or a cloud server.

[0123] The core task of the central signal processing and intelligent decision unit is to receive and process the time-series multi-modal data generated by the precise synchronization control and high-speed data acquisition module. Its processing flow includes sequentially performing feature extraction, intelligent decision, event confirmation and positioning, and finally outputting the accurate diagnosis result of the leakage event.

[0124] In a specific embodiment, the unit logically includes a feature extraction unit, a deep learning model inference unit, and a leakage confirmation and positioning unit. These three units sequentially process data in a pipeline manner.

[0125] The input of the feature extraction unit is time-series multi-modal data First, the unit pre-processes the data, such as filtering and normalization. Then, it extracts multiple dimensions of acousto-optic features from the second harmonic signal S 2f (t) and the multi-path acoustic signal .

[0126] Exemplarily, TDLAS features can include: steady-state concentration indicator values represented by the mean value of the second harmonic signal before being triggered by physical micro-disturbances, and dynamic change characteristics represented by the peak value, rise time, decay constant, etc. of the signal after the disturbance.

[0127] Exemplarily, acoustic features can include: root mean square value, kurtosis, mel frequency cepstral coefficient (MFCC), etc. of each path acoustic signal.

[0128] Further, the unit calculates acoustic array features. It calculates the cross-correlation function R between different microphone channel signals and finds the delay τ corresponding to the peak value of R ij (τ), to determine the arrival time difference τ ij .

[0129]

[0130] Based on multiple sets of arrival time differences and combined with the physical model of formula (5), the spatial position of the sound source can be initially solved.

[0131] In addition, the unit also extracts acousto-optic synchronism and correlation features, such as calculating the time delay and shape cross-correlation coefficient between the second harmonic signal and the acoustic signal envelope, to quantify the causal correlation strength between them. Finally, all the extracted numerical features are combined into a high-dimensional fusion feature vector F i .

[0132] The deep learning model inference unit receives the fusion feature vector F i as input. Inside this unit, a pre-trained deep neural network model is deployed. This model maps the input fusion feature vector to an output probability through its internal complex non-linear transformation.

[0133] In one embodiment, this model utilizes its classification output layer, e.g. a Softmax layer, to output a posterior probability distribution of the scan point belonging to a pre-defined leak state category, e.g. “leak” or “no leak”. This probability distribution is the leak state probability P i .

[0134] P i = DNN(F i ) = [p(no leak), p(leak)] i (8) ;

[0135] where DNN represents the transformation function of the deep neural network model.

[0136] The leak confirmation and localization unit receives the leak state probability P i and the associated raw features of all scan points. First, this unit filters out potential leak points based on a pre-set probability threshold.

[0137] Subsequently, this unit performs physical rule based confirmation on these potential leak points. It examines the candidate events against pre-set criteria of acoustic- optical time synchronicity, acoustic-optical shape correlation and acoustic-optical spatial consistency. For example, it examines whether the time delay between the acoustic event and the optical response is within a physically reasonable range, and whether the acoustic source localization result is consistent with the current scan point location.

[0138] Only when the leak state probability of a potential leak point is significant, and its acoustic-optical features satisfy the above multiple physical consistency criteria, this leak event is finally confirmed. Upon confirmation, this unit can utilize the acoustic array feature computed acoustic source location to refine the leak source localization, and based on the amplitude of the second harmonic signal to preliminarily quantify the leak severity, thus completing the entire detection and diagnosis process.

[0139] The valve volatile organic compound leak laser detection method described below can be mutually corresponding with the valve volatile organic compound leak laser detection system described above.

[0140] Please refer to the accompanying Figure 6 , a valve volatile organic compound leak laser detection method, comprising the following steps:

[0141] S1, scanning and synchronization perturbation data acquisition step: control the detection laser beam of the main detection TDLAS module to scan the target area of the valve point by point, and when the detection laser beam reaches each scanning point, synchronously trigger the laser-induced micro-perturbation module to apply controllable physical micro-perturbation, and synchronously collect the second harmonic signal related to the concentration of volatile organic compounds and the acoustic signal captured by the multi-channel acoustic perception array module by the precise synchronization control and high-speed data acquisition module, forming the time sequence multi-modal data of each scanning point;

[0142] S2, after obtaining the time sequence multi-modal data of each scanning point, then performing signal preprocessing and feature extraction step on the time sequence multi-modal data: preprocessing the time sequence multi-modal data, and extracting TDLAS features, acoustic features, and sound-light synchronization and correlation features, to construct a fusion feature vector for each scanning point;

[0143] S3, after constructing a fusion feature vector for each scanning point, then using the fusion feature vector for the leakage intelligent decision step: inputting the fusion feature vector into the pre-trained deep learning model to output the leakage state probability of each scanning point;

[0144] S4, after obtaining the leakage state probability of each scanning point, and according to the leakage state probability, performing the leakage event confirmation and positioning step: confirming the leakage event according to the preset criterion, and refining the location and preliminarily quantifying the leakage degree of the confirmed leakage source;

[0145] S5, system adaptive adjustment and optimization step: dynamically adjusting the perturbation parameters of the laser-induced micro-perturbation module and the scanning path or scanning point density of the main detection TDLAS module according to the leakage state probability obtained in the current round of detection and the confirmed leakage event;

[0146] S6, after completing the system adaptive adjustment and optimization step, then repeating steps S1 to S5, or finally outputting the leakage diagnosis result when the preset detection condition is met.

[0147] In step S1, synchronously collecting the second harmonic signal related to the concentration of volatile organic compounds and the acoustic signal captured by the multi-channel acoustic perception array module includes: synchronously recording the time sequence of the second harmonic signal and the multi-channel original time sequence of the acoustic signal within the preset time window before, during and after the physical micro-perturbation event trigger, and adding accurate time stamps to all collected data streams.

[0148] In step S2, constructing a fusion feature vector for each scanning point includes: extracting TDLAS features, acoustic features, and sound-light synchronization and correlation features, wherein:

[0149] The TDLAS features include pre-perturbation steady-state concentration indicator, perturbation-induced signal dynamic change feature and frequency domain feature;

[0150] The acoustic features include time-domain features, frequency-domain features, mel-frequency cepstral coefficients of the single-channel acoustic signal, and time-difference-of-arrival and sound source localization parameters based on the multi-channel acoustic signal;

[0151] The acoustic-optical synchronism and correlation features include a time delay and a shape cross-correlation coefficient between the TDLAS response features and the acoustic event features.

[0152] In the S3 step, the step of inputting the fusion feature vector into the pre-trained deep learning model and outputting a leakage state probability includes:

[0153] The parallel multi-modal feature encoding modules inside the deep learning model are used to preliminarily encode the features of different modalities;

[0154] The attention mechanism or splicing operation inside the deep learning model is used to fuse the encoded multi-modal features;

[0155] The classification output layer of the deep learning model is used to output a posterior probability distribution of each scanning point belonging to a predefined leakage state category;

[0156] The leakage state probability is the posterior probability distribution.

[0157] The device of the embodiment can be used to execute the method embodiments, and the principles and technical effects are similar, and thus will not be described herein.

[0158] Although the embodiments of the present application have been shown and described, it can be understood by those of ordinary skill in the art that various changes, modifications, replacements and variations can be made to the embodiments without departing from the principles and spirits of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A valve volatile organic compound leak laser detection system, comprising: The method comprises the following steps: a main detection TDLAS module is used to generate a modulated detection laser beam to scan a target area of a valve, and receive the detection laser beam after passing through a gas area to be detected to obtain a second harmonic signal related to a volatile organic compound concentration; a laser-induced micro-perturbation module is used to synchronously apply controllable physical micro-perturbations to the scanning points or their adjacent accurate positions when the detection laser beam reaches each scanning point; a multi-channel acoustic sensing array module is used to synchronously capture acoustic signals directly generated by valve leakage or induced by the physical micro-perturbations; a precise synchronous control and high-speed data acquisition module is used to accurately control the scanning of the main detection TDLAS module, synchronously trigger the laser-induced micro-perturbation module, and synchronously acquire the second harmonic signal and the acoustic signal to form time-series multi-modal data; a central signal processing and intelligent decision unit is used to process the time-series multi-modal data, extract acoustic-optical features, fuse the acoustic-optical features to make intelligent decisions, confirm and locate the leakage events.

2. The valve volatile organic compound leak laser detection system of claim 1, wherein, The main detection TDLAS module comprises: a tunable diode laser is used to generate the detection laser beam; a two-dimensional scanning galvanometer system is used to drive the detection laser beam to scan the target area of the valve; a photodetector is used to receive the detection laser beam after passing through the gas area to be detected; a lock-in amplifier is used to demodulate the signal received by the photodetector to extract the second harmonic signal related to the volatile organic compound concentration.

3. The valve volatile organic compound leak laser detection system of claim 1, wherein, The laser-induced micro-perturbation module comprises: an auxiliary pulsed laser is used to generate laser pulses according to a trigger signal; a micro-gas nozzle is used to jet micro-gas flow pulses; a high-speed precise electromagnetic valve is used to control the micro-gas nozzle to jet micro-gas flow pulses according to a trigger signal; wherein the laser pulses and the micro-gas flow pulses are used as the controllable physical micro-perturbations; a perturbation energy guiding unit is used to accurately guide the laser pulses or the micro-gas flow pulses to target points.

4. The valve volatile organic compound leak laser detection system of claim 3, wherein, The multi-channel acoustic sensing array module comprises: at least three microphones are used to synchronously pick up the acoustic signals; a plurality of low-noise preamplifiers are respectively connected with the microphones to amplify the picked-up acoustic signals; a plurality of anti-aliasing filters are respectively connected with the low-noise preamplifiers to filter the amplified acoustic signals.

5. The valve volatile organic compound leak laser detection system of claim 1, wherein, The central signal processing and intelligent decision unit comprises: a feature extraction unit is used to extract acoustic-optical features from the time-series multi-modal data, wherein the acoustic-optical features comprise TDLAS features related to the volatile organic compound concentration, time-domain and frequency-domain features of the acoustic signals, acoustic array features, and acoustic-optical synchronism and correlation features, and a fusion feature vector is constructed based on the acoustic-optical features; a deep learning model inference unit is used to receive the fusion feature vector, and output a leakage state probability of each scanning point through a pre-trained deep neural network model. A leakage confirmation and positioning unit is configured to confirm a leakage event and refine a leakage source position based on the leakage state probability and preset criteria of acoustic-optical time synchronization, acoustic-optical shape correlation and acoustic-optical spatial consistency.

6. A method of detecting leaks of volatile organic compounds from valves using laser light, characterized in that The valve volatile organic compound leakage laser detection system according to any one of claims 1-5 comprises the following steps: S1, a scanning and synchronous perturbation data acquisition step: a detection laser beam of a main detection TDLAS module is controlled to perform point-by-point scanning on a target area of a valve, a controllable physical micro-perturbation is applied by a laser-induced micro-perturbation module when the detection laser beam reaches each scanning point, and a precise synchronous control and high-speed data acquisition module is used to synchronously acquire a second harmonic signal related to the volatile organic compound concentration and an acoustic signal captured by a multi-channel acoustic sensing array module, thereby forming time-series multi-modal data of each scanning point; S2, after the time-series multi-modal data of each scanning point is obtained, a signal preprocessing and feature extraction step is performed on the time-series multi-modal data: the time-series multi-modal data is preprocessed, TDLAS features, acoustic features, and acoustic-optical synchronization and correlation features are extracted, and a fusion feature vector is constructed for each scanning point; S3, after the fusion feature vector is constructed for each scanning point, the fusion feature vector is used for a leakage intelligent decision step: the fusion feature vector is input into a pre-trained deep learning model, and a leakage state probability of each scanning point is output; S4, after the leakage state probability of each scanning point is obtained, a leakage event confirmation and positioning step is performed according to the leakage state probability: a leakage event is confirmed by combining a preset criterion, and a confirmed leakage source is refined and a leakage degree is preliminarily quantified.

7. The method of claim 6, wherein the valve is a butterfly valve. In the S1 step, the synchronous acquisition of the second harmonic signal related to the volatile organic compound concentration and the acoustic signal captured by the multi-channel acoustic sensing array module comprises: before, during and after the physical micro-perturbation event is triggered, a preset time window is set, the time series of the second harmonic signal and the multi-channel original time series of the acoustic signal are synchronously recorded, and accurate time stamps are added to all collected data streams.

8. The method of claim 6, wherein the valve is a valve used in a fuel vapor system. In the S2 step, the fusion feature vector is constructed for each scanning point, which comprises: the TDLAS features, the acoustic features, and the acoustic-optical synchronization and correlation features are extracted, wherein: the TDLAS features comprise a pre-perturbation steady-state concentration indicator, a perturbation-induced signal dynamic change feature and a frequency domain feature; the acoustic features comprise time domain features, frequency domain features, mel-frequency cepstral coefficients of a single-channel acoustic signal, and arrival time difference and sound source positioning parameters based on multi-channel acoustic signals; the acoustic-optical synchronization and correlation features comprise a time delay between TDLAS response features and acoustic event features and a shape cross-correlation coefficient.

9. The method of claim 6, wherein the valve is a volatile organic compound leak laser detection method. In the S3 step, the fusion feature vector is input into a pre-trained deep learning model, and the leakage state probability is output, which comprises: a parallel multi-modal feature encoding module inside the deep learning model is used to preliminarily encode features of different modalities; The encoded multi-modal features are fused through an attention mechanism or a concatenation operation inside the deep learning model; A classification output layer of the deep learning model is used to output a posterior probability distribution of each scanning point belonging to a predefined leakage state category; The leakage state probability is the posterior probability distribution.

10. The method of claim 6, wherein the valve is a volatile organic compound leak laser detection method. After the S4 step is performed, the method further includes: S5, system adaptive adjustment and optimization step: according to the leakage state probability obtained in the current round of detection and the confirmed leakage event, dynamically adjusting the perturbation parameters of the laser-induced micro-perturbation module and the scanning path or scanning point density of the main detection TDLAS module; S6, after the system adaptive adjustment and optimization step is completed, the steps S1 to S5 are repeatedly executed, or the final leakage diagnosis result is output when the preset detection condition is met.

Images