Intelligent testing device for waterproof grade of electronic throttle valve

Abstract

The invention relates to the technical field of nondestructive testing of automobile parts, and discloses an intelligent testing device for the waterproof grade of an electronic throttle valve. The device comprises an acoustic phased array transducer, a laser Doppler vibration meter, a cavity ring-down spectroscopy gas analyzer and a data processing unit. The unit excites a measured piece through acoustics, and synchronously collects vibration response measured by a high-precision laser Doppler vibration meter and trace tracer gas leakage data monitored by a cavity ring-down spectroscopy gas analyzer. After the data is processed, a vibration leakage coupling heat map for locating leakage and a leakage path fingerprint for characterizing characteristics can be generated. Vibration response data and tracer gas leakage data caused by acoustic excitation are synchronously collected, high-precision non-contact vibration measurement is achieved through a laser Doppler vibration meter, trace gas is monitored in real time through a cavity ring-down spectroscopy gas analyzer, and the multi-source data processing capacity of an intelligent model is combined. Evaluation of the waterproof performance of the electronic throttle valve is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of non-destructive testing of automobile parts, in particular to an intelligent testing device for the waterproof level of an electronic throttle valve. BACKGROUND

[0002] As the core control unit of the engine intake system of a vehicle, an electronic throttle valve integrates precise electronic circuits and sensing elements inside. In the complex environment of actual vehicle operation, the electronic throttle valve must have high sealing and waterproof capabilities to prevent water from entering and causing circuit short circuits, signal abnormalities, and even functional failures, thereby ensuring the stable operation of the engine and driving safety. Therefore, it is crucial to perform strict and reliable waterproof performance testing on each electronic throttle valve product before it is shipped.

[0003] Currently, the methods commonly used for waterproof or air-tightness testing of industrial products include the water immersion bubble method, pressure drop method, and artificial tracer gas sniffing method. However, these traditional techniques have obvious limitations when applied to high-precision mechatronic components such as electronic throttle valves. These methods often rely on subjective human observation or can only provide a general assessment of the overall leakage rate, making it difficult to achieve objective, accurate, and repeatable quantification of the waterproof level.

[0004] More importantly, when a product is found to be unqualified, these existing technologies often cannot accurately indicate the specific location of the leak. A small leak can be caused by local defects in the shell joint surface, slight deformation of the sealing ring, or structural defects at a specific injection point. Knowing only that there is a leak at the macro level provides very limited information for improving production processes and diagnosing the root cause of defects. In addition, some testing methods lack sufficient sensitivity to detect small leaks that pose a potential threat to the long-term reliability of the product, while other methods can cause invasive effects on the tested part itself, even risking damage to internal electronic components, which greatly reduces testing efficiency and reliability. SUMMARY

[0005] To address the shortcomings of existing technologies, the present application provides an intelligent testing device for the waterproof level of an electronic throttle valve, which solves the problem that traditional techniques can only provide general pass / fail judgments and cannot objectively and accurately quantify the waterproof performance.

[0006] To achieve the above purpose, the present application is implemented by the following technical solution: an intelligent testing device for the waterproof level of an electronic throttle valve, comprising:

[0007] a detection box, a machine box is fixedly connected to the side wall of the detection box, and a synchronous control and data processing unit is arranged inside;

[0008] A fixture is arranged inside the detection box for fixing the electronic throttle valve;

[0009] An acoustic phased array transducer is electrically connected to the synchronous control and data processing unit and located at one side of the fixture;

[0010] A tracer gas supply unit has its output end connected to the internal cavity of the electronic throttle valve fixed on the fixture;

[0011] A laser Doppler vibrometer has an optical path towards the fixture and sends the collected vibration response data to the synchronous control and data processing unit;

[0012] A cavity ring-down spectroscopy gas analyzer is connected to the internal cavity of the detection box through a sampling tube and sends the collected leakage data to the synchronous control and data processing unit inside the detection box;

[0013] The synchronous control and data processing unit is used to synchronously control the excitation of the acoustic phased array transducer and the data collection of the laser Doppler vibrometer and the cavity ring-down spectroscopy gas analyzer, and runs an intelligent model to process the vibration response data and the leakage data.

[0014] Preferably, the tracer gas is filled in the sealed cavity inside the electronic throttle valve;

[0015] An acoustic excitation is applied to the electronic throttle valve;

[0016] The vibration response data of the electronic throttle valve shell and the leakage data of the tracer gas in the environment around the electronic throttle valve caused by the acoustic excitation are synchronously collected;

[0017] The vibration response data and the leakage data are input into a pre-trained intelligent model, and a quantitative waterproof level representing the waterproof performance of the electronic throttle valve is obtained by processing the intelligent model.

[0018] Preferably, the quantitative waterproof level includes:

[0019] A waterproof performance index representing the degree of waterproof performance; or a diagnostic report containing information of potential leakage source position.

[0020] Preferably, before the step of applying the acoustic excitation, further comprising:

[0021] A wideband acoustic excitation is applied to the electronic throttle valve, and its global vibration response spectrum and reference leakage curve are collected, and the global vibration response spectrum and the reference leakage curve are part of the data input into the intelligent model.

[0022] Preferably, the acoustic excitation is a dynamic modulation signal, further comprising:

[0023] generate a leakage path fingerprint based on cross-correlation between the dynamic modulation signal and a time function of the leakage data;

[0024] wherein the intelligent model takes the leakage path fingerprint as one of inputs for determining the quantified waterproof level.

[0025] Preferably, the acoustic excitation is a focused sound beam, further comprising:

[0026] controlling the focused sound beam to scan the surface of the electronic throttle valve shell;

[0027] synchronously collecting local vibration response data and corresponding leakage data at the scanning point position during the scanning process, and determining incremental leakage data according to comparison between the leakage data and a reference leakage value;

[0028] generating a vibration-leakage coupling heat map based on the local vibration response data and the incremental leakage data;

[0029] wherein the intelligent model takes the vibration-leakage coupling heat map as one of inputs for determining the quantified waterproof level.

[0030] Preferably, the step of generating the vibration-leakage coupling heat map specifically comprises:

[0031] calculating a leakage contribution factor of the scanning point position according to the incremental leakage data at the scanning point position and the sound power or sound pressure amplitude applied at the point;

[0032] and generating the vibration-leakage coupling heat map based on the leakage contribution factors of all scanning points.

[0033] Preferably, the focused sound beam is generated by a dynamic modulation signal; further comprising:

[0034] generate a leakage path fingerprint based on cross-correlation between the dynamic modulation signal and a time function of the leakage data collected during the scanning process;

[0035] wherein the intelligent model determines the quantified waterproof level based on the leakage path fingerprint and the vibration-leakage coupling heat map.

[0036] Preferably, the step of synchronous collection specifically comprises:

[0037] collecting the vibration response data in a non-contact manner by the laser Doppler vibrometer;

[0038] and simultaneously monitoring real-time trace gas concentration changes by a cavity ring-down spectroscopy gas analyzer to obtain the leakage data.

[0039] Preferably, the intelligent model is pre-trained through a machine learning process comprising the following steps:

[0040] Obtain the vibration response data and leakage data of a plurality of electronic throttle valve samples as training inputs;

[0041] Obtain the known waterproof level corresponding to the samples as training labels;

[0042] And establish a mapping relationship between the training inputs and the training labels through a training algorithm.

[0043] The present application provides an electronic throttle valve waterproof level intelligent testing device. It has the following advantages:

[0044] 1. The present application synchronously collects vibration response data and tracer gas leakage data induced by acoustic excitation, and uses a laser Doppler vibration meter to realize high-precision non-contact vibration measurement and a cavity ring-down spectroscopy gas analyzer to realize real-time monitoring of trace gases. Combined with the processing capability of the intelligent model for multi-source data, it realizes objective and high-precision quantitative evaluation of the waterproof performance of the electronic throttle valve, and can provide a fine waterproof performance index.

[0045] 2. The present application uses an acoustic phased array transducer to apply a focused acoustic beam to scan the surface of the electronic throttle valve shell, synchronously collects local vibration response data and incremental leakage data at the scanning points, and generates a vibration leakage coupling heat map based on this, so as to accurately identify and locate potential leakage source positions in a non-contact and non-destructive manner, avoiding the limitations of traditional destructive detection or blind search.

[0046] 3. The present application integrates a synchronous control and data processing unit and a pre-trained intelligent model. The intelligent model can process complex vibration response data, leakage data, leakage path fingerprints and vibration leakage coupling heat maps, and automatically output quantitative waterproof levels or diagnostic reports, significantly improving the automation level of the detection process and the consistency of the evaluation results. BRIEF DESCRIPTION OF DRAWINGS

[0047] Figure 1 is a perspective view of the present application;

[0048] Figure 2 is a cross-sectional view of the detection box of the present application;

[0049] Figure 3 is a flowchart of the electronic throttle valve waterproof level intelligent testing method of the present application;

[0050] Figure 4 is a flowchart of the intelligent model training and construction of the present application.

[0051] Wherein, 1, detection box; 2, machine box; 3, clamp; 4, acoustic phased array transducer; 5, tracer gas supply unit; 6, laser Doppler vibrometer; 7, cavity ring-down spectroscopy gas analyzer. DETAILED DESCRIPTION

[0052] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the specification of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0053] Embodiment one:

[0054] Referring to the drawings Figure 1 And Figure 2 An electronic throttle valve waterproof grade intelligent testing device, which comprises a detection box 1, a machine box 2, a clamp 3, an acoustic phased array transducer 4, a tracer gas supply unit 5, a laser Doppler vibrometer 6 and a cavity ring-down spectroscopy gas analyzer 7.

[0055] The detection box 1 is used to provide a relatively closed test environment to accommodate the electronic throttle valve to be tested and to collect the tracer gas leaked therefrom. The machine box 2 is fixedly connected to the side wall of the detection box 1, and a synchronous control and data processing unit is arranged in the machine box 2. The unit is the control and calculation core of the whole device, responsible for generating control timing, collecting multi-source data and executing data processing algorithms.

[0056] The clamp 3 is arranged in the inside of the detection box 1, and is designed to stably clamp electronic throttle valves of different models, so as to ensure that the position and posture of the electronic throttle valves remain constant during the test.

[0057] The tracer gas supply unit 5 comprises a tracer gas source, a pressure regulating valve and a connecting pipeline, and the output end thereof is connected to the internal cavity of the electronic throttle valve installed on the clamp 3 through a sealing interface, for filling the internal cavity with tracer gas of a predetermined pressure and concentration. The tracer gas can be sulfur hexafluoride or helium and other inert gases with extremely low content in standard atmosphere.

[0058] The acoustic phased array transducer 4 is arranged in the detection box 1, on one side of the clamp 3, and is electrically connected with the synchronous control and data processing unit in the machine box 2. The transducer is composed of multiple independent piezoelectric transducer array elements, and the synchronous control and data processing unit can independently control the amplitude and phase of the excitation signal transmitted to each array element, and can generate various forms of acoustic field through the principle of beam synthesis. For example, a wideband acoustic excitation can be generated by applying the same wideband random signal to all array elements; a specific time sequence characteristic acoustic excitation can be generated by applying a dynamically modulated signal; and through precise calculation and phase delay applied to each array element, the acoustic energy can be converged into a focused acoustic beam, and the phase delay can be changed through electrical control to make the focused acoustic beam scan the surface of the electronic throttle valve shell on the predetermined path.

[0059] The laser Doppler vibrometer 6 is arranged outside or inside the detection box 1, and its optical path is directed towards the clamp 3, so that the laser beam emitted by it can irradiate the surface of the shell of the electronic throttle valve to be measured. Based on the Doppler effect, the instrument non-contactly obtains the vibration velocity or displacement of the irradiation point in the normal direction of the shell by measuring the frequency shift of the reflected laser, thereby obtaining the vibration response data. The laser Doppler vibrometer 6 sends the collected vibration response data to the synchronous control and data processing unit in the machine box 2 in real time.

[0060] The cavity ring-down spectroscopy gas analyzer 7 is connected to the inner cavity of the detection box 1 through a sampling tube, and is used to continuously or periodically extract the gas sample in the detection box 1. The analyzer uses the cavity ring-down spectroscopy technology to invert the concentration of specific tracer gas molecules by measuring the ring-down time of laser pulses in a high-precision optical resonant cavity containing a gas sample. This technology has high detection sensitivity and can monitor the trace changes in the concentration of tracer gas in the detection box 1 caused by leakage in real time, and use the concentration change data as leakage data, and send it to the synchronous control and data processing unit in the machine box 2.

[0061] The synchronous control and data processing unit in the machine box 2 can be specifically composed of a field programmable gate array, a digital signal processor, a high-performance embedded computer or a combination thereof. The unit performs the following functions:

[0062] I. Generate and output an electrical signal to drive the acoustic phased array transducer 4;

[0063] II. Generate a unified synchronous clock signal to ensure that the data acquisition of the laser Doppler vibrometer 6 and the data acquisition of the cavity ring-down spectroscopy gas analyzer 7 are strictly aligned in time sequence;

[0064] III. Receive and store the vibration response data and leakage data from the two measuring instruments 6 and 7;

[0065] Four, run a pre-trained intelligent model, the collected data processing and analysis, and finally output representing the electronic throttle valve waterproof performance of the quantitative level.

[0066] Embodiment two:

[0067] Referring to the drawings Figure 3 An electronic throttle valve waterproof level intelligent testing method, the method is applied to the electronic throttle valve waterproof level intelligent testing device, and can include the following steps:

[0068] 2.1, basic test flow

[0069] S1, filling the sealed cavity inside the electronic throttle valve with tracer gas.

[0070] This step is realized by the tracer gas supply unit 5. The tracer gas supply unit 5 delivers the tracer gas with a predetermined pressure and concentration to the internal sealed cavity of the electronic throttle valve fixed on the clamp 3 through the connecting pipeline. The filling process ensures that there is a pressure difference between the inside and outside of the electronic throttle valve, so as to promote the diffusion of the tracer gas outward when there is a small leakage channel.

[0071] S2, applying acoustic excitation to the electronic throttle valve.

[0072] This step is realized by the acoustic phased array transducer 4. The synchronous control and data processing unit inside the cabinet 2 sends a control signal to the acoustic phased array transducer 4, so that it generates and emits sound waves to the surface of the shell of the electronic throttle valve. The applied acoustic excitation can make the shell of the electronic throttle valve vibrate. The type, frequency range and spatial distribution of the acoustic excitation can be selected according to the specific test mode, such as wideband acoustic excitation, dynamic modulation signal or focused acoustic beam.

[0073] S3, synchronously collecting the vibration response data of the shell of the electronic throttle valve and the leakage data of the tracer gas in the environment around the electronic throttle valve caused by the acoustic excitation.

[0074] This step is the core sensing link of the method, which is realized by the laser Doppler vibrometer 6 and the cavity ring-down spectroscopy gas analyzer 7, and is coordinated by the synchronous control and data processing unit.

[0075] Specifically, the vibration response data of the surface of the shell of the electronic throttle valve is collected in a non-contact manner by the laser Doppler vibrometer 6. The laser Doppler vibrometer 6 emits a laser beam to the surface of the shell of the electronic throttle valve, and measures the vibration velocity or displacement of the point according to the frequency change of the reflected light, so as to obtain the vibration response data thereof. The data can represent the dynamic deformation characteristics of the electronic throttle valve under acoustic excitation, especially the vibration response characteristics of the small cracks and joint mismatching areas of the shell.

[0076] Meanwhile, the trace gas concentration in the chamber of the detection box 1 is monitored in real time by the cavity ring-down spectroscopy gas analyzer 7 to obtain leakage data. The cavity ring-down spectroscopy gas analyzer 7 extracts gas from the detection box 1 through a sampling pipe and detects the concentration of the trace gas therein by using high-sensitivity optical technology. If the electronic throttle valve leaks, the trace gas inside will enter the detection box 1 through the leakage channel, causing the concentration of the trace gas in the detection box 1 to increase. The real-time monitoring capability of the cavity ring-down spectroscopy gas analyzer 7 can capture this weak and dynamic concentration change as leakage data.

[0077] The synchronization control and data processing unit is responsible for generating a unified time reference to ensure that the data acquisition of the laser Doppler vibrometer 6 and the cavity ring-down spectroscopy gas analyzer 7 are accurately aligned on the time axis, thereby establishing the time correlation between acoustic excitation, vibration response, and gas leakage.

[0078] S4, input the vibration response data and the leakage data into a pre-trained intelligent model, and obtain a quantitative waterproof level representing the waterproof performance of the electronic throttle valve by processing the intelligent model.

[0079] This step is performed in the synchronization control and data processing unit inside the machine case 2. The synchronization control and data processing unit inputs the vibration response data and the leakage data collected in S3 into a pre-stored and running intelligent model. The intelligent model has established a mapping relationship from these multi-source physical measurement data to the waterproof performance level of the electronic throttle valve through a large amount of sample data training. Specifically, this mapping relationship can be fixed by the internal parameters of the trained model. In the inference stage, the calculation process of the model can be exemplarily described as follows:

[0080] First, the branches of the model extract features from the preprocessed input data:

[0081] ① The convolutional neural network branch processes the input vibration leakage coupling heat map and outputs an image feature vector .

[0082] ② The recurrent neural network branch processes the input leakage path fingerprint and outputs a sequence feature vector .

[0083] ③ The fully connected network branch processes the input vector composed of the global vibration response spectrum and the reference leakage curve and outputs a global feature vector .

[0084] Then, the fusion layer splices these feature vectors from different modalities to form a comprehensive feature vector :

[0085] ;

[0086] Finally, the comprehensive feature vector is calculated to obtain the final quantitative waterproof level . The calculation process can be divided into two cases according to the type of task:

[0087] Case one: waterproof level is a continuous numerical value

[0088] At this time, the output layer is a linear unit, and its calculation relationship is:

[0089] ;

[0090] wherein, is the weight matrix of the output layer, is the bias vector, and both are fixed parameters obtained after model training. The output is a continuous numerical index, which can be defined in an interval of 0 to 100, for example. The numerical index is positively correlated with the waterproof performance, for example, it can be preset that > 90 is excellent, 70≤ ≤ 90 is qualified, and 70< X < 70 is unqualified, so as to realize fine grade evaluation.

[0091] Case two: waterproof level is a discrete category At this time, a Softmax activation function will be connected after the output layer to output the probability that the sample belongs to each predefined level. Assuming that the predefined waterproof levels are L1: qualified, L2: critical, and L3: unqualified, the model will output a probability distribution vector . The calculation formula of each probability is:

[0092] ;

[0093] wherein, is the total number of categories (here, 3), is the linear output of the output layer to the th category (i.e. ). Finally, the model takes the category with the highest probability as the prediction result :

[0094] ;

[0095] The result directly indicates the waterproof level of the electronic throttle valve.

[0096] 2.2 Mode one: global rapid evaluation mode

[0097] ​In this mode, the step of applying acoustic excitation (S2) is preceded by a preliminary step of pre-scanning, i.e. applying a broadband acoustic excitation to the electronic throttle valve. The broadband excitation is generated by the acoustic phased array transducer 4 under the control of the synchronous control and data processing unit, with a frequency range covering the natural frequency range of the main structural components of the electronic throttle valve. Under this excitation, the laser Doppler vibrometer 6 collects the global vibration response of the surface of the electronic throttle valve housing, and the global vibration response spectrum is obtained by Fourier transform and other signal processing methods. At the same time, the cavity ring-down spectroscopy gas analyzer 7 collects the corresponding baseline leakage data to form a baseline leakage curve. The global vibration response spectrum and the baseline leakage curve are transmitted to the intelligent model as the basis data for subsequent evaluation, reflecting the inherent vibration characteristics and initial state of the leakage of the test piece.

[0098] Subsequently, the excitation and collection steps for global evaluation are formally entered. At this time, the step of applying acoustic excitation (S2) is specifically to apply a dynamic modulation signal. The signal is generated by the synchronous control and data processing unit, which can be a pseudo-random sequence signal or a linear frequency modulation signal, and its time waveform has unique and identifiable autocorrelation characteristics.

[0099] At the same time of applying the dynamic modulation signal, the synchronous collection step (S3) is started. The laser Doppler vibrometer 6 collects vibration response data, and the cavity ring-down spectroscopy gas analyzer 7 collects leakage data. Due to the effect of acoustic vibration coupling, if there is a leakage channel, the characteristics of the applied dynamic modulation acoustic signal will modulate the gas flow through the leakage channel, thereby leaving a mark related to the excitation signal in the time function of the leakage data collected by the cavity ring-down spectroscopy gas analyzer 7.

[0100] In this mode, the processing step (S4) further includes generating a leakage path fingerprint. Specifically, the synchronous control and data processing unit performs cross-correlation operation on the time function of the dynamic modulation excitation signal and the time function of the leakage data to generate a cross-correlation function :

[0101] ;

[0102] wherein:

[0103] is the time function of the dynamic modulation signal;

[0104] is the time function of the leakage data collected by the cavity ring-down spectroscopy gas analyzer 7;

[0105] is time;

[0106] is time delay.

[0107] Due to the dynamic modulation signal has a good autocorrelation, which has a correlation function with the leakage signal modulated by it will present one or more significant peaks. The waveform of the cross-correlation function constitutes a leakage path fingerprint. The fingerprint reflects the transfer characteristics of the entire path from the acoustic excitation point to the gas leakage and detected by the sensor, such as transmission delay and attenuation characteristics. Finally, the intelligent model will use this leakage path fingerprint, together with the previously collected global vibration response spectrum and the reference leakage curve, as one of the comprehensive inputs to determine the quantitative waterproof level that characterizes the overall waterproof performance.

[0108] 2.3 Mode two: high-precision leak source positioning mode

[0109] In this mode, the step of applying acoustic excitation (S2) is specifically to apply a focused acoustic beam. The acoustic phased array transducer 4, under the precise control of the synchronous control and data processing unit, forms a highly concentrated acoustic focal point, i.e. a focused acoustic beam, by adjusting the excitation phase and amplitude of each array element. The focused acoustic beam can concentrate acoustic energy on a small area on the surface of the electronic throttle valve shell.

[0110] The method further includes controlling the focused acoustic beam to scan the surface of the electronic throttle valve shell. The synchronous control and data processing unit pre-sets a scanning path, such as a grid-shaped path, and by continuously adjusting the array element phase of the acoustic phased array transducer 4, the focused acoustic beam is moved point by point or region by region on the predetermined surface area of the electronic throttle valve shell at a pre-set step size and speed.

[0111] During the scanning process, the synchronous acquisition step (S3) is performed synchronously at each scanning point or scanning area. When the focused acoustic beam is focused on a specific position on the surface of the shell, the laser Doppler vibrometer 6 collects the local vibration response data at that scanning point position. At the same time, the cavity ring-down spectroscopy gas analyzer 7 monitors and collects the corresponding leakage data in real time. The synchronous control and data processing unit compares these local vibration response data and leakage data with the pre-set reference leakage value to determine the incremental leakage data at that scanning point position. The reference leakage value can be the average of the leakage data under no acoustic excitation or non-focused acoustic excitation. The incremental leakage data reflects the amount of leakage additionally induced due to the enhancement of local vibration after applying the focused acoustic beam at a specific point.

[0112] Based on the collected local vibration response data and incremental leakage data, the processing step (S4) further includes generating a vibration-leakage coupling heat map. Specifically, for each scanning point , the synchronous control and data processing unit determines the color value of the corresponding pixel in the heat map according to the incremental leakage data at that point and the sound power or sound pressure amplitude applied at the point , the leakage contribution factor at the point is calculated . Its calculation relationship can be expressed as:

[0113] ;

[0114] wherein:

[0115] represents the incremental leakage data at the scanning point , and

[0116] represents the sound power or sound pressure amplitude of the focused sound beam applied at the scanning point .

[0117] A specific and commonly used function form is:

[0118] ;

[0119] By calculating the leakage contribution factor at all scanning points , the synchronous control and data processing unit obtains a two-dimensional leakage contribution factor matrix. Subsequently, the unit performs spatial interpolation and pseudo-color coding processing on the matrix to generate a visual vibration leakage coupling heat map. In the heat map, different colors or brightness intensities correspond to different leakage contribution factor values, and high leakage contribution factor areas appear as hot spots on the heat map, indicating the location of the leakage defect on the surface of the electronic throttle valve shell. The intelligent model takes this vibration leakage coupling heat map as one of its inputs to determine a diagnostic report containing potential leakage source location information, or as an important basis for quantifying the waterproof level.

[0120] 2.4 Compound diagnosis mode

[0121] In this compound diagnosis mode, the driving signal for generating the focused sound beam is a dynamic modulation signal. Specifically, the synchronous control and data processing unit generates a dynamic modulation signal with specific time sequence characteristics and provides the signal as an excitation source to the acoustic phased array transducer 4. Based on this signal, the acoustic phased array transducer 4 adjusts the phase delay of each array element to synthesize a sound beam that is spatially focused and temporally modulated.

[0122] This method then performs a scanning process similar to the high-precision leakage source positioning mode. That is, the dynamic modulated focused sound beam is controlled to scan the surface of the electronic throttle valve shell. During the scanning process, the synchronous acquisition step (S3) acquires the incremental leakage data at each scanning point ​the local vibration response data of the point, and the real-time leakage data time function acquired by the cavity ring-down spectroscopy gas analyzer 7 .

[0123] The processing step (S4) performs a dual analysis task in this mode. First, it performs the same steps as in the high-precision leak source localization mode to generate a vibration leak coupling heat map. That is, by calculating the incremental leakage data and the acoustic power at each scanning point , the leakage contribution factor is obtained, and a vibration leak coupling heat map is generated based on the matrix of all scanning points to determine the spatial location of one or more potential leak sources.

[0124] Secondly, the processing step also performs the generation of a leak path fingerprint for the data collected during the scanning process. Specifically, for the regions identified as high leak risk on the vibration leak coupling heat map or all scanning points, the synchronous control and data processing unit extracts the leakage data time function corresponding to the point, and generates a leak path fingerprint specific to the leak point based on the cross-correlation between the original dynamic modulation signal and the leakage data time function .

[0125] Finally, the intelligent model takes the vibration leak coupling heat map and the leak path fingerprint corresponding to the high-risk regions on the heat map as its composite input. The intelligent model determines a more detailed quantitative waterproof level based on the spatial location information of the leak source provided by the vibration leak coupling heat map, and in combination with the leak channel transmission characteristic information provided by the leak path fingerprint. The level can include a comprehensive diagnostic report composed of the coordinates of the leak point, the characteristic parameters of the leak channel, etc.

[0126] Embodiment Three: Training and Construction of Intelligent Model

[0127] Referring to the accompanying drawings Figure 4 , Figure 4 is a flowchart of the training and construction of the intelligent model according to an embodiment of the present application. The intelligent model is pre-trained through a machine learning process including the following steps:

[0128] S10, Constructing a training data set

[0129] This step provides basic data for model training. This step specifically includes:

[0130] First, a sufficient number of electronic throttle valve physical samples with diversity are obtained. These samples should systematically cover different states, including: samples that fully meet the production standards, critical samples whose performance is at the boundary of qualified and unqualified, and samples carrying pre-set or known defects. Defect samples can be artificially manufactured to simulate various failure modes that occur in actual use, such as different degrees of aging of the sealing ring material, micron-level cracks in the shell at specific locations, and assembly deviations in the sealing structure of the connector interface.

[0131] Second, for each of the above physical samples, the electronic throttle valve waterproof level intelligent testing device of the present application is used to perform complete data acquisition. According to the pre-set test program, one or more test modes are applied to the sample, such as the global rapid evaluation mode and the high-precision leakage source positioning mode. In this process, the synchronous control and data processing unit completely records a series of measurement data associated with the sample, which will constitute the training input in the training data set. The specific data collected can include: global vibration response spectrum, reference leakage curve, leakage data time series collected under dynamic modulation signal excitation, leakage path fingerprint generated from the time series, and vibration leakage coupling heat map generated in the focused sound beam scanning mode.

[0132] Third, the true waterproof performance of each physical sample is calibrated by an independent, recognized authoritative test method to obtain its corresponding training label. The authoritative test method can be the standard immersion test defined in the international protection level certification, such as immersing the sample in water of a specified depth for a specified time, and then determining whether it fails due to water immersion. Through this test, each sample is assigned an exact waterproof level.

[0133] Finally, the training input data collected from each physical sample is paired with its training label calibrated by the authoritative test method. The collection of all these data pairs collectively constitutes the structured final training data set used for subsequent model training.

[0134] S20, pre-processing the collected data

[0135] The purpose is to convert the original, heterogeneous data collected from the electronic throttle valve samples in step S10 into a unified, standardized format suitable for input into the intelligent model, to improve the training efficiency and prediction performance of the model. This step specifically includes the following processing operations:

[0136] Firstly, the raw data of various types are standardized. For numerical data, such as the amplitude of the global vibration response spectrum, the concentration value of the baseline leakage curve, the cross-correlation coefficient of the leakage path fingerprint, and the leakage contribution factor of the vibration leakage coupling heat map, normalization or standardization methods are used for scaling. For example, Z-score standardization can be used, and the calculation relationship is:

[0137] ;

[0138]

[0139] is the standardized data point;

[0140] is the original data point;

[0141] and are the mean and standard deviation of the feature over the entire training dataset, respectively;

[0142] This method makes the data conform to the standard normal distribution with a mean of 0 and a variance of 1.

[0143] Secondly, noise suppression and feature extraction are performed on time series data and spectral data. For the original vibration signal collected by the laser Doppler vibration meter 6 or the original leakage data time series collected by the cavity ring-down spectroscopy gas analyzer 7, a digital filter can be used to remove high-frequency noise or irrelevant frequency components to improve the signal-to-noise ratio. For spectral data, smoothing processing can be performed. From these processed time series data, statistical features or frequency domain features can be further extracted as input features of the intelligent model to more compactly represent the signal characteristics.

[0144] Thirdly, image-type data is processed. For the vibration leakage coupling heat map generated in the scanning mode, its original size and resolution vary due to the scanning parameters. Therefore, size normalization is needed to adjust all heat maps to the unified image size preset by the intelligent model. In addition, the pixel values of the heat map can be remapped to enhance the contrast of the defect area and converted to the input format required by the model, such as a grayscale image matrix or a multi-channel image.

[0145] Finally, all pre-processed data of different modalities are integrated and packaged to form a structured multi-modal input. For example, normalized spectral data, extracted time series features, and normalized size heat map matrices can be combined into a single complex input vector or tensor that can be received by the intelligent model. This integration process ensures that the model can simultaneously utilize multi-dimensional information from different sensors for comprehensive judgment.

[0146] S30, training, verification and deployment of the model​

[0147] This step aims to build a computational model capable of accurately predicting the waterproofing level from input data using the dataset prepared in step S20 through a machine learning algorithm. This step specifically includes:

[0148] First, an intelligent model structure is selected and constructed. According to the multi-modal characteristics of the input data, a fusion neural network model can be constructed. The model structure can include:

[0149] a convolutional neural network branch for processing two-dimensional image data such as vibration leakage coupling heat maps;

[0150] a recurrent neural network or its variants such as long short-term memory network branch for processing sequence data such as leakage path fingerprints;

[0151] and a fully connected network branch for processing one-dimensional vector data such as global vibration response spectrum. Each branch extracts high-dimensional features from different types of data, which are then connected and integrated by a fusion layer, and finally output the prediction results through one or more fully connected layers.

[0152] Second, the training process of the model is performed. The preprocessed training dataset in step S20 is input into the constructed model. In each iteration, the model performs forward propagation calculation on the input data to obtain a predicted waterproofing level. Then, a predefined loss function is used to calculate the error between the predicted value and the corresponding true label in the training dataset. For regression tasks where the waterproofing level is a continuous numerical value, the mean square error can be selected as the loss function, and for classification tasks where the waterproofing level is a discrete category, the cross-entropy loss can be selected, with the calculation relationship being:

[0153] For regression tasks, the calculation relationship of the mean square error is:

[0154] ;

[0155] where:

[0156] is the number of training samples in this batch;

[0157] is the true waterproofing level value of the th sample;

[0158] is the waterproofing level value predicted by the model for this sample;

[0159] is the index of the sample.

[0160] For classification tasks, the calculation relationship of the cross-entropy loss is:

[0161]

[0162] wherein:

[0163] is the number of batch samples;

[0164] is the total number of classes;

[0165] is a symbolic function;

[0166] is the probability of the model predicting the sample belongs to the class .

[0167] Again, according to the calculated loss, an optimization algorithm is used to update the model parameters. The optimization algorithm is a mathematical method used to systematically adjust the internal parameters of the model to minimize the loss function, and is the core engine driving the model to learn. Its basic principle is the gradient descent method, that is, by calculating the gradient of the loss function with respect to the model parameters, and adjusting the parameters in the opposite direction of the gradient.

[0168] In a specific embodiment, the optimization algorithm is the Adam optimizer, which is a highly efficient adaptive learning rate optimization algorithm. In the th iteration of training, its parameter update process follows the following calculation relationship:

[0169] 1. Calculate the gradient of the current batch data:

[0170] ;

[0171] wherein,

[0172] represents the gradient of the loss function with respect to the model parameter at the th iteration;

[0173] represents the model parameter at the end of the last iteration.

[0174] 2. Update the biased first moment estimate (momentum term):

[0175] ;

[0176] wherein,

[0177] is the first moment estimate of the gradient (exponential moving average of the gradient);

[0178] is the exponential decay rate hyperparameter for the first moment.

[0179] 3. Update the biased second moment estimate (adaptive learning rate term):

[0180] ;

[0181] where,

[0182] is the second moment estimate of the gradient (exponentially moving average of the squared gradient), recording the magnitude of the historical gradients;

[0183] is the exponential decay rate hyperparameter for the second moment.

[0184] 4. Compute the bias-corrected first and second moment estimates:

[0185] ;

[0186] ;

[0187] This step aims to correct the bias towards zero caused by the initializations of and to zero at the beginning of training.

[0188] 5. Perform the final parameter update:

[0189] ;

[0190] where,

[0191] is the updated model parameter;

[0192] is the initial learning rate;

[0193] is a small constant added to prevent division by zero.

[0194] This training process is repeated for multiple epochs over the entire training dataset, with the model parameters being continuously fine-tuned through the iterative calculations of the above optimization algorithm, until the performance metrics on an independent validation set reach a state of convergence or satisfy the pre-set performance threshold

[0195] Finally, the model is solidified and deployed. After the training process is completed, the model parameter combination with the best performance on the validation set is saved and solidified. The solidified intelligent model is deployed to the synchronous control and data processing unit inside the cabinet 2. In actual test applications, this unit will call the deployed model to perform real-time processing and analysis on newly collected electronic throttle data and directly output the quantified waterproof level or diagnostic report.

Claims

1. An electronic throttle valve waterproof grade intelligent testing device, characterized in that, The utility model relates to a kind of electronic throttle valve waterproof performance testing system, including: Detection box (1), the detection box (1) side wall fixedly connected with machine case (2), and inside is provided with synchronous control and data processing unit; Clamp (3), it is arranged in detection box (1) inside, for fixed electronic throttle valve; Acoustic phased array transducer (4), it is electrically connected to the synchronous control and data processing unit, and located in the clamp (3) side; Trace gas supply unit (5), its output end is connected with the internal cavity of the electronic throttle valve fixed on the clamp (3); Laser Doppler Vibrometer (6), it has optical path towards the clamp (3), and vibration response data collected is sent to the synchronous control and data processing unit; Cavity ring-down spectroscopy gas analyzer (7), it is communicated with the inner cavity of the detection box (1) by sampling tube, and leakage data collected is sent to the synchronous control and data processing unit in machine case (2) inside; Wherein, synchronous control and data processing unit are used for synchronous control the excitation of acoustic phased array transducer (4) and the data acquisition of laser Doppler Vibrometer (6) and cavity ring-down spectroscopy gas analyzer (7), and run intelligent model to process vibration response data and leakage data.

2. An electronic throttle valve waterproof grade intelligent testing method applied to the electronic throttle valve waterproof grade intelligent testing device of any one of claim 1, characterized in that, Including the following steps: Fill trace gas in the sealed cavity inside electronic throttle valve; Apply acoustic excitation to electronic throttle valve; Synchronously collect vibration response data of electronic throttle valve shell caused by the acoustic excitation and leakage data of the trace gas in the environment around the electronic throttle valve; The vibration response data and the leakage data are input into a pre-trained intelligent model, and the quantitative waterproof grade representing the waterproof performance of the electronic throttle valve is obtained by the intelligent model.

3. The electronic throttle valve waterproof grade intelligent test method according to claim 2, characterized in that, The quantitative waterproof grade includes: Waterproof performance index representing the degree of waterproof performance, or diagnostic report containing potential leakage source location information.

4. The electronic throttle valve waterproof grade intelligent test method according to claim 2, characterized in that, Before the step of applying acoustic excitation, further including: Apply wideband acoustic excitation to the electronic throttle valve, collect its global vibration response spectrum and reference leakage curve, and input the global vibration response spectrum and the reference leakage curve as part of the data input into the intelligent model.

5. The electronic throttle valve waterproof grade intelligent test method according to claim 2, characterized in that, The acoustic excitation is a dynamic modulation signal, and further including: Based on the cross-correlation between the dynamic modulation signal and the time function of the leakage data, generate a leakage path fingerprint; Wherein, the intelligent model takes the leakage path fingerprint as one of the inputs for determining the quantitative waterproof grade.

6. The electronic throttle valve waterproof grade intelligent test method according to claim 2, characterized in that, The acoustic excitation is a focused acoustic beam, and further including: Control the focused acoustic beam to scan the surface of the electronic throttle valve shell; During scanning, synchronously collect local vibration response data at the scanning point position and corresponding leakage data, and determine incremental leakage data according to the comparison between the leakage data and the reference leakage value; Based on the local vibration response data and the incremental leakage data, generate a vibration leakage coupling heat map; Wherein, the intelligent model takes the vibration leakage coupling heat map as one of the inputs for determining the quantitative waterproof grade.

7. The electronic throttle valve waterproof grade intelligent test method according to claim 6, characterized in that, The step of generating a vibration leakage coupling heat map specifically includes: calculating a leakage contribution factor of the point according to the incremental leakage data at the position of the point and the sound power or sound pressure amplitude applied at the point; generating the vibration leakage coupling heat map based on the leakage contribution factors of all scanning points.

8. The electronic throttle valve waterproof grade intelligent test method according to claim 6, characterized in that, The focused sound beam is driven by a dynamic modulation signal; further comprising: generating a leakage path fingerprint based on the cross-correlation between the dynamic modulation signal and the time function of the leakage data collected during the scanning process; wherein the intelligent model determines the quantitative waterproof level jointly according to the leakage path fingerprint and the vibration leakage coupling heat map.

9. The electronic throttle valve waterproof grade intelligent test method according to claim 2, characterized in that, The step of synchronously collecting specifically comprises: collecting the vibration response data in a non-contact manner by a laser Doppler vibrometer; and simultaneously monitoring the trace gas concentration change in real time by a cavity ring-down spectroscopy gas analyzer to obtain the leakage data.

10. The electronic throttle valve waterproof grade intelligent test method according to claim 2, characterized in that, The intelligent model is obtained by a machine learning process comprising the following steps: obtaining the vibration response data and leakage data of a plurality of electronic throttle valve samples as training input; obtaining the known waterproof level corresponding to the samples as training label; and establishing the mapping relationship between the training input and the training label by a training algorithm.

Images