Liquid information detection method and system based on ultrasonic technology

Abstract

The invention discloses a liquid information detection method and system based on an ultrasonic technology. The method comprises the following steps: firstly, acquiring ultrasonic multiple echo signals under a preset working condition; the peak information, the time domain characteristic, the waveform characteristic, the phase characteristic and other information of the echo are obtained; secondly, establishing a machine learning model, and inputting feature information such as echo peak voltage, time delay and phase into the machine learning model for training; obtaining a trained machine learning model; and finally, inputting an actually detected ultrasonic echo signal into the trained machine learning model for processing to obtain the liquid information distribution condition of the detected target. According to the method, liquid component identification and high-precision detection of liquid information of different liquid levels are realized through array ultrasonic transducer and echo signal analysis, echo signal characteristics can be quickly mapped to liquid attribute information in combination with a machine learning model, and excellent intelligent processing capacity is shown. The method is high in real-time performance and applicability, can adapt to different liquid types and densities and various environmental conditions, and meets the detection requirements in an industrial environment and the application scene of higher-precision detection.

Description

Technical Field

[0001] The present invention relates to the technical field of ultrasonic detection, and in particular to a method and system for detecting liquid information based on ultrasonic technology. Background Art

[0002] Liquid detection technology plays a key role in many fields such as industry, environment, and energy, and is widely used in scenarios such as liquid storage, chemical reactions, liquid level control, and environmental monitoring. The accuracy, stability, and real-time nature of liquid level information are crucial for the safety and efficiency of equipment operation. Especially in complex liquid storage or transportation systems, the liquid level height and distribution information have irreplaceable value for liquid management, operation control, and equipment fault prevention.

[0003] Currently, significant progress has been made in liquid level detection technology, which is widely used in various storage tanks and containers. However, there are still certain challenges for non-contact detection technology of mixed stratified liquids. Mixed stratified liquids are usually composed of liquids with different densities and refractive indices, and their interfaces may not be obvious, and the physical properties of the liquids may change over time, increasing the complexity of detection. Nevertheless, some studies have attempted to solve this problem. For example, a non-contact liquid level detection method based on a multi-layer pulse neural network realizes the detection of the liquid level by analyzing the dynamic changes of the liquid surface in the video image. However, this method requires a bright light environment and has obvious defects for detecting liquids in closed containers. In addition, a method for implementing non-contact liquid level measurement using a reflectometer chip indirectly infers the liquid level change by measuring the impedance change of the transmission line. However, these methods mainly target single liquids or liquids with obvious interfaces, and the non-contact detection technology for mixed stratified liquids is still in the exploratory stage and has not yet formed a mature technical system. Therefore, developing a non-contact detection technology applicable to mixed stratified liquids has good application prospects and practical value.

[0004] Due to its advantages such as non-contact measurement, fast response, and high precision, ultrasonic technology has been widely used in the field of liquid detection in recent years. Since the acoustic impedances of different liquids are different, the reflection coefficients at their interfaces are also different, and non-contact detection of liquids is achieved by processing the reflected echoes. Traditional ultrasonic detection technology usually uses a single sensor that combines transmitting and receiving or two sensors, one for transmitting and one for receiving, for detection. The detection principle is to determine the position of the liquid surface by the propagation time or reflection characteristics of the acoustic wave signal. However, this method can only obtain the position information of the liquid surface and cannot obtain the composition information of the liquid. Therefore, its application scope and functions are greatly limited. In the complex scenario of detecting a closed container filled with multiple layers of liquids with different compositions, it is necessary to simultaneously obtain the stratification situation of the liquid surface in the closed container and the liquid composition information, and the existing detection methods and technologies using a single sensor cannot detect the multi-layer liquid surface detection or liquid composition detection. Therefore, there is an urgent need for new detection methods and technologies.

[0005] Therefore, a method and system capable of simultaneously detecting the distribution information of multiple liquid levels and the information of different liquid components in a container are required. Summary of the Invention

[0006] In view of this, the purpose of the present invention is to provide a method and system for detecting liquid information based on ultrasonic technology, which uses non-contact detection of liquid components and liquid level distribution based on ultrasonic technology.

[0007] To achieve the above purpose, the present invention provides the following technical solutions:

[0008] The method for detecting liquid information based on ultrasonic technology provided by the present invention includes the following steps:

[0009] S1: Obtain the ultrasonic echo signal of the preset working condition;

[0010] S2: Process the echo signal to obtain information such as echo time-domain characteristics and waveform characteristics;

[0011] S3: Establish a machine learning model, and use information such as peak information, time-domain characteristics, waveform characteristics, and phase characteristics extracted from the echo signal as feature data input into the machine learning model for training;

[0012] S4: Obtain the trained machine learning model, and input the preprocessed ultrasonic echo signal into the machine learning model for training to obtain the trained machine learning model;

[0013] S5: Input the actually detected ultrasonic echo signal into the trained machine learning model for processing to obtain the liquid information data distribution of the detected target.

[0014] Furthermore, it further includes the following steps:

[0015] S6: Establish an echo and liquid information database, and the echo and liquid information database is used to establish the association information between the echo signal and the liquid information.

[0016] Furthermore, the liquid information data includes liquid category information or / and liquid level information;

[0017] The liquid category information is obtained by comparing the constructed echo and liquid information database to obtain liquid feature information;

[0018] The liquid level information is determined by the position of the externally set pulse ultrasonic signal transmitting device.

[0019] Furthermore, the echo and liquid information database is established according to the correspondence between the acquired liquid information data and the liquid types; by comparing the echo and liquid information database with the acquired liquid information, the corresponding liquid type can be obtained; or

[0020] The echo and liquid information database determines the liquid level position according to the position of the pulsed ultrasonic signal transmitting device; by comparing the echo and liquid information database with the position of the pulsed ultrasonic signal transmitting device, the corresponding liquid level position is obtained.

[0021] Furthermore, the pulsed ultrasonic signal device is arranged in any one of the following ways:

[0022] Linear array: The pulsed ultrasonic signal transmitting devices are arranged at intervals along a straight line;

[0023] Circular array: The pulsed ultrasonic signal transmitting devices are distributed in a circular or elliptical manner;

[0024] Matrix array: The pulsed ultrasonic signal transmitting devices are distributed in a two-dimensional grid arrangement;

[0025] Irregular array: The pulsed ultrasonic signal transmitting devices are distributed according to the container shape and detection requirements;

[0026] Multi-layer array: The pulsed ultrasonic signal transmitting devices are arranged in a multi-layer array structure on the same vertical plane.

[0027] Furthermore, the machine learning model uses a neural network; the input layer of the neural network is the voltage amplitude of the echo signal, and the output layer is the liquid information data; or

[0028] The machine learning model includes, but is not limited to, any one of support vector machine (SVM), random forest (RF), backpropagation neural network (BP), convolutional neural network (CNN), and hybrid model.

[0029] Furthermore, the pulsed ultrasonic signals are emitted at intervals, and the interval time satisfies the following relationship:

[0030]

[0031] where T is the pulse interval time; n1 is the number of times of receiving the pulsed ultrasonic echo; and v is the propagation speed of the sound wave in the container wall material.

[0032] Furthermore, the pulsed ultrasonic signal contains at least two cycles, and its emission frequency satisfies the following relationship:

[0033]

[0034] Wherein, n2 is the number of cycles of the pulsed ultrasonic signal, and n2 ≥ 2. f0 represents the emission frequency of the pulsed ultrasonic signal and the resonance frequency of the piezoelectric transducer. v represents the sound velocity of the sound wave in the container wall material, and d represents the container wall thickness.

[0035] The liquid information detection system based on ultrasonic technology provided by the present invention includes an ultrasonic probe, a signal emission control module, an acquisition control module, and a signal processing module.

[0036] The ultrasonic probe is used to emit a pulsed ultrasonic signal to a preset position on the container wall; the wavelength of the pulsed ultrasonic signal is less than the container wall thickness.

[0037] The signal emission control module is used to control the emission process of the ultrasonic signal.

[0038] The acquisition control module is used to control the acquisition process of the echo signal.

[0039] The signal processing module is used to obtain liquid information data in the container through the analysis and processing of the ultrasonic echo signal; the analysis and processing are carried out through a machine learning model, that is, the characteristic data of the echo signal is input into the machine learning model, and the liquid information data representing the liquid information is obtained through the analysis and processing of the machine learning model; the liquid information data is compared with the constructed echo and liquid information database to obtain liquid characteristic information, and the liquid characteristic information includes liquid category information.

[0040] Furthermore, the neural network adopts a BP neural network; the BP neural network includes an input layer, a hidden layer, and an output layer; the input layer is used to input the voltage amplitude of the echo signal; the output layer is used to output liquid information data, and the liquid information data includes liquid category information or / and liquid liquid level information; the BP neural network is trained using a standard data set with known liquid physical parameters, and the standard data set includes the echo amplitude of the liquid and the corresponding liquid types. The data that has not been used for training is input into the trained BP neural network to detect different liquid types; the BP neural network is trained in the following manner:

[0041] First, a single echo training model is performed. The input layer of the BP neural network is used to input the single echo amplitude A1 for training.

[0042] Then, a double echo training model is performed. The input layer of the BP neural network is used to input the single echo amplitude A1 and the double echo amplitude A2 for training.

[0043] Then, a triple echo training model is performed. The input layer of the BP neural network is used to input the triple echo amplitudes A1, A2, and A3 for training; as the trained BP neural network.

[0044] Finally, perform n - time echo training on the model. The input layer of the BP neural network is used to input the amplitudes A1, A2, A3 of the n - time echoes and A n for training; serve as the trained BP neural network.

[0045] The BP neural network can be trained separately according to the amplitude A1 of the first - time echo, the amplitude A2 of the second - time echo, the amplitude A3 of the third - time echo, or the amplitude A of the n - time echo. n Train separately.

[0046] The BP neural network is gradually trained through any combination of the amplitudes A1 of the first - time echo, the amplitudes A2 of the second - time echo, the amplitudes A3 of the third - time echo, or the amplitudes A of the n - time echo for training the model. n Train step by step.

[0047] Furthermore, the ultrasonic echo signal includes the amplitude and / or the waveform signal of the echo signal.

[0048] The beneficial effects of the present invention are as follows:

[0049] The method and system for liquid information detection based on ultrasonic technology provided by the present invention first obtain the ultrasonic multi - time echo signals under preset working conditions; process the multi - time echo signals to obtain information such as time - domain characteristics and waveform characteristics; then establish a machine - learning model, and use the peak information, time - domain characteristics, waveform characteristics, phase characteristics, etc. extracted from the echo signals as feature data input into the machine - learning model for training; obtain the trained machine - learning model. By inputting the pre - processed ultrasonic echo signals into the machine - learning model for training, a trained machine - learning model is obtained; finally, input the actually detected ultrasonic echo signals into the trained machine - learning model for processing to obtain the liquid information data distribution of the detected target.

[0050] This method realizes high - precision detection of liquid component identification and liquid information at different liquid levels through array layout and analysis of n - time echo signals. Combining with the BP neural network model, the echo signal features can be quickly mapped to liquid attribute information, showing excellent intelligent processing capabilities. It has strong real - time performance and applicability, can adapt to different liquid types, densities, and various environmental conditions, and meets the detection requirements in industrial environments. At the same time, the device adopts a modular design, has good scalability, and can flexibly expand the array scale to adapt to application scenarios of larger containers or higher - precision detection.

[0051] Other advantages, objectives, and features of the present invention will be described to some extent in the subsequent specification, and to some extent, will be obvious to those skilled in the art based on the study of the following text, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the following specification. Description of the Drawings

[0052] To make the objectives, technical solutions and beneficial effects of the present invention clearer, the present invention provides the following attached drawings for illustration.

[0053] Figure 1 It is a schematic flow diagram of the method for identifying liquid components and detecting the liquid level distribution.

[0054] Figure 2 It is a structural diagram of the device for identifying liquid components and detecting the liquid level distribution.

[0055] Figure 3 It is a schematic diagram of the array probe setting.

[0056] Figure 4 It is a schematic flow diagram of detecting liquid information and liquid level distribution.

[0057] Figure 5 It is a schematic diagram of the classification detection results of multiple liquids using a neural network model.

[0058] Figure 6 It is a table of the detection results of liquid types. Detailed implementation manners

[0059] The following further describes the present invention in conjunction with the attached drawings and specific embodiments, so that those skilled in the art can better understand the present invention and be able to implement it, but the cited embodiments do not limit the present invention.

[0060] Embodiment 1

[0061] As Figure 1 shown, the liquid information detection method and system based on ultrasonic technology provided in this embodiment are used for identifying liquid components and detecting the liquid level distribution, and can detect liquid components non-contact and in real time, and can accurately obtain the distribution information of liquids at different heights, overcoming the problems of low liquid level detection accuracy and easy damage of equipment in the prior art. The specific steps are as follows:

[0062] S1: Obtain the ultrasonic multiple echo signals under the preset working conditions;

[0063] S2: Process the multiple echo signals to obtain information such as peak value information, time domain characteristics, waveform characteristics, and phase characteristics;

[0064] S3: Establish a machine learning model, and use the time domain characteristics, waveform characteristics, etc. extracted from the echo signals as the feature data input into the machine learning model for training; this feature data can reflect liquid information, and the machine learning model uses a neural network;

[0065] S4: Obtain the trained machine learning model. By inputting the preprocessed ultrasonic echo signals into the machine learning model for training, a trained machine learning model is obtained. The ultrasonic echo signals are the echo signals reflected by the detection target after being emitted by the array ultrasonic transducer;

[0066] S5: Input the actually detected ultrasonic echo signals into the trained machine learning model for processing to obtain the liquid information data distribution of the detected target. In this embodiment, the array ultrasonic echo signals can be used to quickly obtain the liquid information data, so as to determine the distribution of the liquid level;

[0067] S6: Establish an echo and liquid information database, which is used to establish the association information between the echo signals and the liquid information; In this embodiment, the echo signals are collected for various known liquids, then the characteristic data of the echo signals are extracted, and the association information is established according to the liquid information data and the characteristic data of the known liquids to obtain the echo and liquid information database of the liquid. Finally, the algorithm model can be trained using the echo and liquid information database to obtain a trained model with liquid characteristics, and then the echo information of the unknown liquid collected is identified to obtain the liquid information data. The liquid information data includes liquid category information and may also include liquid level information;

[0068] In this embodiment, the echo signals are used as characteristic data and input into the machine learning model, and the liquid information data is obtained through analysis and identification processing by the machine learning model; An echo and liquid information database is constructed according to the liquid information data;

[0069] The echo and liquid information database of this embodiment is established according to the corresponding relationship between the obtained liquid information data and the liquid types; By comparing the echo and liquid information database with the obtained liquid information, the corresponding liquid types can be obtained;

[0070] A one-to-one correspondence is established between the liquid information data and the echo signals; Therefore, through this echo and liquid information database, the corresponding liquid information data, that is, the liquid type information, can be queried according to the actual echo signals;

[0071] The liquid information data also includes liquid level information; The liquid level information is determined by the position of the externally arranged pulsed ultrasonic signal transmitting device;

[0072] The echo and liquid information database of this embodiment can also be established according to the one-to-one correspondence between the positions of the pulsed ultrasonic signal transmitting devices; By querying the echo and liquid information database, the corresponding liquid level information can be obtained according to the position of the pulsed ultrasonic signal transmitting device;

[0073] The machine learning model of this embodiment adopts a neural network; the input layer of the neural network is the voltage amplitude of the echo signal, and the output layer is the liquid information; a standard data set of existing liquid physical parameters is selected to train the neural network, and the training set contains the characteristic data of several liquids and the corresponding output liquid information; the echo amplitude of the data not used for training is input into the trained BP neural network to detect different liquid types.

[0074] In this embodiment, the ultrasonic echo signal is a signal carrying peak information, time-domain characteristics, waveform characteristics, phase characteristics, etc.

[0075] In this embodiment, the ultrasonic transducer emits a pulsed ultrasonic signal with a preset frequency to the container wall, and then records the corresponding echo signal; samples and stores the echo signal emitted each time; the liquid in the container in this embodiment can be different types of liquids, and the boundary between different liquids represents different types of liquids; there are different interaction relationships between different types of liquids and the container wall. When the ultrasonic wave signal passes through the container wall and is reflected from the liquid surface in contact with the container wall to form an echo signal; the echo signal carries data on the liquid-container wall gap information; when ultrasonic waves propagate in a medium and encounter the interface of different media, reflection and refraction will occur. Among them, the intensity and time delay of the reflected signal are related to the acoustic impedance and propagation path of the medium, and the acoustic impedances of different media are different.

[0076] As Figure 2 shown, the basic principle of the liquid information detection method provided in this embodiment: in liquid detection, the characteristics of pulsed ultrasonic echoes are significantly affected by the reflection coefficient at the liquid-container wall interface, and the reflection coefficient is determined by the acoustic impedance of the liquid and the acoustic impedance of the container wall material. It overcomes the deficiency of only being able to identify the liquid level in the traditional method and can also identify the specific liquid type information.

[0077] In addition, during the propagation of sound waves in the container wall, they are affected by scattering and attenuation characteristics, which further reduces the resolution ability of the reflected signal, especially significantly in a high-noise environment. On the other hand, although the array probe can provide rich spatial data, phase information, amplitude information, nonlinear information, etc. in liquid distribution detection, the data it collects is huge and the information content is complex, and it cannot be directly deduced or calculated by existing acoustic theoretical equations.

[0078] Finally, the liquid detection signal is analyzed and processed by a signal processing method of machine learning. Using a neural network and combining with the time-series echo signals collected by the array ultrasonic probe, it can not only accurately identify the tiny differences between the liquid and interface echoes, but also effectively process a large amount of high-dimensional data and eliminate the interference of environmental noise. In terms of pattern recognition and non-linear data fitting, the neural network has powerful learning and generalization abilities, can automatically extract the deep features in the signal, thereby improving the recognition accuracy of the detection system for complex liquid distributions. In addition, by analyzing the spatio-temporal data with the neural network, the distribution of the liquid at different positions can be accurately obtained, and at the same time, the intelligent classification of the liquid type can be realized, overcoming the deficiencies of the existing acoustic theory methods in identifying liquid components.

[0079] Therefore, combining the echo signal with the neural network provides an effective solution for the high-precision and non-contact detection of liquid distribution. This method can not only give full play to the spatial acquisition ability of the array probe, but also utilize deep learning technology to optimize the processing and analysis of complex signals, showing excellent robustness and detection efficiency in complex environments, providing a new research direction for liquid surface distribution detection and liquid type identification.

[0080] The duration of the pulsed ultrasonic signal in this embodiment is two cycles; the frequency of the pulsed signal further satisfies the following relationship:

[0081]

[0082] where n2 is the number of cycles of the pulsed ultrasonic signal, and n2≥2, f0 represents the emission frequency of the pulsed ultrasonic signal and the resonant frequency of the piezoelectric transducer, v represents the sound velocity of the sound wave in the container wall material, d represents the container wall thickness; λ represents the wavelength of the pulsed ultrasonic signal; the thickness is selected to suppress the standing wave effect, and the formation of the standing wave needs to satisfy: d = kλ / 2 (k is an integer); when λ << d, it is difficult to form a stable standing wave mode in the container wall. At this time, the multiple reflections of the sound wave in the wall show random phase superposition, rather than fixed node enhancement, which can avoid the periodic modulation of the echo amplitude by the standing wave and ensure the linear response relationship between the signal amplitude and the liquid characteristics.

[0083] To avoid the overlap of the nth pulsed echo and the main wave of the next emitted pulse, the pulse interval T should ensure that all echoes generated by the previous pulse are completely received, including the multiple reflection echoes. The pulse time interval needs to satisfy:

[0084]

[0085] where T is the pulse interval time; n1 is the number of times of receiving the pulsed ultrasonic echo; v is the propagation speed (sound velocity) of the sound wave in the container wall material.

[0086] Frequency Selection and Arrangement of the Array Ultrasonic Probe in this Embodiment: The array ultrasonic probe is used for non-contact detection of the liquid surface distribution and liquid characteristics;

[0087] In terms of probe frequency selection, to avoid aliasing of the multiple reflection echoes from the container wall, the wavelength of the pulsed ultrasound should satisfy: λ << d, to reduce the standing wave effect; avoid subsequent pulse interference with the unreceived echoes; and enable the echo signal to include liquid information;

[0088] In terms of the array arrangement, the array ultrasonic probe is composed of several ultrasonic transducers with both transmitting and receiving functions, and is arranged outside the container tank body. According to different detection requirements and container shapes, it can be designed into various arrangement structures, including but not limited to:

[0089] Linear array: The transducers are arranged at equal intervals or specific intervals along a straight line, suitable for regular container shapes or linear detection paths;

[0090] Circular array: The transducers are distributed in a circular or elliptical manner, suitable for the detection of centrally symmetric liquids;

[0091] Matrix array: The transducers are arranged in a two-dimensional grid, suitable for large-range detection covering the entire container surface;

[0092] Irregular array: For complex container shapes and specific detection requirements, the transducers are arranged in an irregular manner to optimize the coverage range;

[0093] Multi-layer array: A multi-layer array structure is arranged on the same vertical plane for multi-dimensional information acquisition of complex three-dimensional liquid surface distributions.

[0094] Each ultrasonic transducer realizes signal transmission and reception during operation, and measures the liquid levels at different heights and liquid information at different positions respectively. Through the flexible array structure design, this method is not only applicable to a variety of complex liquid environments, but also can meet the diverse requirements of liquid surface distribution detection and liquid property identification.

[0095] This method uses an array ultrasonic probe to emit ultrasonic pulse waves, and multi-point collects the multiple echo signals inside the container wall, and can obtain ultrasonic echo signals from multiple different positions at the same time, obtain the specific information of the liquid, and then reconstruct the liquid surface distribution map.

[0096] Pulsed Ultrasonic Signal Transmission and Multiple Echo Acquisition in this Embodiment:

[0097] Each probe periodically emits pulsed ultrasonic signals to the container wall, and the signals are reflected at the interface between the container wall and the liquid;

[0098] Multiple echo signals:

[0099] The initial signal is emitted from the interface between the container wall and the probe, and the signal reflected back after reaching the interface between the container wall and the liquid is called echo 1; the signal of echo 1 is reflected again by the interface between the container wall and the probe, and the signal reflected back after reaching the interface between the container wall and the liquid again is called echo 2; and so on. The signal of echo n is reflected again by the interface between the container wall and the probe, reaches the interface between the container wall and the liquid and is reflected back, which is called echo n+1;

[0100] Feature extraction of the echo signal in this embodiment:

[0101] Digitize and preprocess the collected echo signal to extract characteristic parameters:

[0102] The peak voltage of the echo signal, which is used to reflect the interface reflection coefficient and is related to the acoustic impedance of the liquid;

[0103] The time delay of the echo signal, which is used to calculate the echo propagation path and is related to the thickness of the container wall;

[0104] The phase change and frequency components of the echo signal, which are used to describe the signal characteristics in a complex liquid environment.

[0105] Application of multiple neural network models in this embodiment:

[0106] 1. Model type and training

[0107] According to different working condition characteristics, the following multiple neural network models are used to process the echo signal and infer the liquid properties:

[0108] BP neural network: As the basic model, it is used to analyze the signal feature-liquid information mapping in a simple and stable environment;

[0109] CNN (Convolutional Neural Network): For the frequency domain characteristics of the echo signal, convolutional operations are used to extract high-dimensional data features, which performs excellently in a high-noise environment;

[0110] RNN (Recurrent Neural Network) and LSTM: Utilize the time series characteristics of the echo signal to learn the multi-level signal dynamic changes and adapt to the dynamic detection of complex liquid levels;

[0111] Hybrid model: Combining the characteristics of CNN and RNN, integrating the signal spatial features and time features, providing more comprehensive and accurate detection capabilities.

[0112] This embodiment uses a BP neural network to analyze and process the echo signal.

[0113] 2. Model training process

[0114] Under the experimental conditions of known liquid levels and liquid types, collect the echo signal characteristics multiple times, and perform model training through the following steps:

[0115] Feature construction: Construct the feature parameters of the n - th echo into a feature vector group to form an input data set;

[0116] Model construction and optimization: According to the requirements of object detection, build BP, CNN, RNN and hybrid network models respectively;

[0117] Use the training data to adjust the network weights and parameters, and optimize the non - linear mapping between features and liquid properties;

[0118] Comprehensive evaluation of multiple models: Compare the performance of multiple networks according to the actual detection environment, and flexibly select a single model or a combined model according to the scenario requirements.

[0119] Measurement and determination of liquid distribution in this embodiment: Use the trained neural network model to process the newly collected echo signals in real time and quickly generate the physical information of liquids at different heights. Specifically, the neural network uses the features of the collected signals (such as time - domain waveform, frequency - domain distribution and amplitude, etc.), combines with the pre - calibrated data set, and automatically calculates the physical parameters related to the target liquid. Through the echo signals obtained by the transducer arrays arranged at different positions, the model can compare the physical information of liquids in different heights and regions, and then realize the comprehensive analysis and dynamic visualization display of the liquid distribution and multi - layer structure in the whole container.

[0120] Embodiment 2

[0121] As Figure 2 shown, this embodiment also provides a liquid information detection system based on ultrasonic technology for liquid composition identification and liquid liquid - level distribution detection device, which specifically includes an ultrasonic probe, a signal emission control module, a collection control module, and a signal processing module;

[0122] The ultrasonic probe is used to emit pulsed ultrasonic signals to a preset position on the container wall; the wavelength of the pulsed ultrasonic signal is less than the thickness of the container wall;

[0123] The signal emission control module is used to control the emission process of the ultrasonic signal;

[0124] The collection control module is used to control the collection process of the echo signal;

[0125] The signal processing module is used to obtain the liquid information in the container through the analysis and processing of the ultrasonic echo signal; the analysis and processing is carried out through a machine learning model, that is, the feature data of the echo signal is input into the machine learning model, and the liquid information data representing the liquid information is obtained through the analysis and processing of the machine learning model; the liquid information data is compared with the constructed echo - liquid information database to obtain the liquid feature information, and the liquid feature information includes liquid category information;

[0126] As Figure 3 shown, Figure 3 is a schematic diagram of the array probe setting. The ultrasonic probe in this embodiment uses an array ultrasonic probe, specifically as follows: Longitudinal array ultrasonic probe: It is composed of multiple ultrasonic transducers that can both transmit and receive, and is arranged outside the container tank body. According to the detection requirements, the array can be designed into various arrangement structures, including but not limited to linear array, circular array, matrix array, irregular array, and multi-layer array. The ultrasonic transducers in the linear array probe emit pulsed ultrasound to the tank body at different height positions of the container, and determine the liquid information based on the reflected echoes. Combine the physical positions of different ultrasonic transducers outside the tank body to determine the distribution level of the liquid in the container.

[0127] Signal processing module: Transmit the required signals, filter and digitize the collected echo signals, and extract the characteristic parameters of the first, second, third, and nth echo signals, including peak voltage, phase change, echo delay, waveform change, etc.

[0128] As Figure 4 shown, the machine learning model in this embodiment uses a neural network; the input layer of the neural network is the voltage amplitude of the echo signal, and the output layer is the liquid information; a standard data set with existing liquid physical parameters is selected to train the neural network, and the training set contains the characteristic data of several liquids and the corresponding output liquid information; input the echo amplitude of the data that has not been used for training into the trained BP neural network to detect different liquid types.

[0129] Neural network module: Pre-install multiple neural network models (including BP neural network, convolutional neural network, long short-term memory network, support vector regression, etc.), which are used to process the extracted echo features and generate the physical information of the liquid. Through the model selection mechanism, according to the detection requirements and the characteristics of the training data, automatically match the optimal neural network model; use the selected model to achieve an accurate mapping from the echo features to the liquid information.

[0130] Liquid distribution analysis module: Based on the liquid physical information output by the neural network module, comprehensively analyze the liquid level and liquid characteristics at different heights; generate the distribution level of the liquid in the container (liquid level height, thickness and properties of different liquid layers), and present it in a visual form (such as a graphical interface or a report). It is applicable to single-layer liquid, multi-layer liquid, and mixed distribution scenarios of multiple liquids, meeting the complex requirements in the industrial environment.

[0131] The liquid information detection system based on ultrasonic technology provided by this embodiment analyzes and processes the echo voltage amplitude through a BP neural network to identify liquid information detection;

[0132] 1. Experimental environment and equipment:

[0133] In this embodiment, a transceiver-integrated ultrasonic transducer is adopted, with a working frequency of 5 MHz, a pulse emission interval of 1 ms, and a single transducer is arranged outside the container; Container and liquid: The container is made of steel with a wall thickness of 0.8 cm, and the measured liquids are 75% ethanol, 95% ethanol, water, air, and rapeseed oil respectively; Signal acquisition device: Use a Tektronix MSO64 oscilloscope with a sampling rate set to 20 MHz to record the ultrasonic three-time echo signals.

[0134] 2. Implementation process:

[0135] Obtaining echo signals: Transmit a 5-MHz pulsed ultrasonic signal to the container wall, with a signal period of 2, and record the three-time echo signals of different liquids. Sample and store the echo signals for each transmission, and repeat the acquisition 160 groups to ensure stability.

[0136] Extracting echo signal characteristics: Process the collected time-domain signals to extract the characteristic points of the echo signals of 5 substances. The voltage amplitudes of the three-time echo signals are A1, A2, and A3 respectively.

[0137] Constructing BP neural network and liquid detection: Design a three-layer BP neural network.

[0138] First, use the first-time echo amplitude A1 as the input to construct a first-time echo training model. Its input layer contains only 1 node (A1), the number of hidden layer nodes is 19, and the output layer node is the liquid type.

[0139] Next, use the first-time echo amplitude A1 and the second-time echo amplitude A2 to construct a second-time echo training model. At this time, the input layer contains 2 nodes (A1, A2), the number of hidden layer nodes is 19, and the output layer node is the liquid type.

[0140] Finally, comprehensively use the three-time echo amplitudes A1, A2, and A3 to construct a complete training model. The input layer has 3 nodes (echo amplitudes A1, A2, A3), the number of hidden layer nodes is 19, and the output layer node is the liquid type.

[0141] All models are trained using a standard data set with known liquid physical parameters. The data set includes the echo amplitudes of the liquids and the corresponding liquid types. Input the data that has not been used for training into the trained BP neural network to detect different liquid types.

[0142] Results and effects: By using the first-time echo, the second-time echo, and the three-time echo signals for training respectively, observe the influence of different echo data sets on the model performance. The liquid type detection results are as Figure 5 and Figure 6As shown. When using the one - echo training model, the classification accuracy rate is 97.9%. When using the two - echo training model, the classification accuracy rate is 99.5%. Finally, the model that comprehensively uses the A1, A2, and A3 three - echo signals reaches a 100% classification accuracy rate. The confusion matrix shows that the model prediction is completely consistent with the true category and there is no misclassification. During the model training process, the mean squared error (MSE) of different data sets (training set, validation set, and test set) shows an obvious downward trend and reaches the optimal at the 13th round, with the mean error being only 1.2322e - 4, indicating that the model has good fitting ability and stability. The error - decreasing curves of the three groups of data are highly consistent, proving that the model has not experienced overfitting or underfitting situations.

[0143] The above - described embodiments are only preferred embodiments given to fully illustrate the present invention, and the protection scope of the present invention is not limited thereto. Equivalent substitutions or transformations made by those skilled in the art on the basis of the present invention are all within the protection scope of the present invention. The protection scope of the present invention is subject to the claims.

Claims

1. A method for detecting liquid information based on ultrasonic technology, characterized in that: It includes the following steps: S1: Obtain the ultrasonic echo signal of the preset working condition; S2: Process the echo signal to obtain information such as echo time-domain characteristics and waveform characteristics; S3: Establish a machine learning model, and use the information such as peak value information, time-domain characteristics, waveform characteristics, and phase characteristics extracted from the echo signal as feature data input into the machine learning model for training; S4: Obtain the trained machine learning model. By inputting the preprocessed ultrasonic echo signal into the machine learning model for training, a trained machine learning model is obtained; S5: Input the actually detected ultrasonic echo signal into the trained machine learning model for processing to obtain the liquid information data distribution of the detected target.

2. The liquid information detection method based on ultrasonic technology according to claim 1, wherein: It also includes the following steps: S6: Establish an echo and liquid information database, which is used to establish the association information between the echo signal and the liquid information.

3. The liquid information detection method based on ultrasonic technology according to claim 1, characterized in that: The liquid information data includes liquid category information and / or liquid liquid level information; The liquid category information obtains liquid characteristic information by comparing with the established echo and liquid information database; The liquid liquid level information is determined by the position of the externally set pulsed ultrasonic signal transmitting device.

4. The liquid information detection method based on ultrasonic technology according to claim 2, characterized in that: The echo and liquid information database is established according to the corresponding relationship between the obtained liquid information data and the liquid types; by comparing the echo and liquid information database with the obtained liquid information, the corresponding liquid types can be obtained; or The echo and liquid information database determines the liquid level position according to the position of the pulsed ultrasonic signal transmitting device; by comparing the echo and liquid information database with the position of the pulsed ultrasonic signal transmitting device, the corresponding liquid level position is obtained.

5. The liquid information detection method based on ultrasonic technology according to claim 1, characterized in that: The pulsed ultrasonic signal device is arranged in any of the following ways: Linear array: The pulsed ultrasonic signal transmitting devices are arranged at intervals along a straight line; Circular array: The pulsed ultrasonic signal transmitting devices are distributed in a circular or elliptical manner; Matrix array: The pulsed ultrasonic signal transmitting devices are distributed in a two-dimensional grid arrangement; Irregular array: The pulsed ultrasonic signal transmitting devices are distributed according to the container shape and detection requirements; Multi-layer array: The pulsed ultrasonic signal transmitting devices are arranged in a multi-layer array structure on the same vertical plane.

6. The liquid information detection method based on ultrasonic technology according to claim 1, characterized in that: The machine learning model uses a neural network; the input layer of the neural network is the voltage amplitude of the echo signal, and the output layer is the liquid information data; or The machine learning model includes, but is not limited to, any one of support vector machine (SVM), random forest (RF), feedforward neural network (BP), convolutional neural network (CNN), and hybrid model.

7. The liquid information detection method based on ultrasonic technology according to claim 1, wherein The pulsed ultrasonic signal is emitted at intervals, and the interval time satisfies the following relationship: where T is the pulse interval time; n1 is the number of times of receiving the pulsed ultrasonic echo; v is the propagation speed of the sound wave in the container wall material.

8. The liquid information detection method based on ultrasonic technology according to claim 1, wherein: The pulsed ultrasonic signal contains at least two cycles, and its emission frequency further satisfies the following relationship: where n2 is the number of cycles of the pulsed ultrasonic signal, and n2≥2, f0 represents the emission frequency of the pulsed ultrasonic signal and the resonant frequency of the piezoelectric transducer, v represents the sound speed of the sound wave in the container wall material, and d represents the container wall thickness.

9. A liquid information detection system based on ultrasonic technology, characterized in that: It includes an ultrasonic probe, a signal emission control module, an acquisition control module, and a signal processing module; The ultrasonic probe is used to emit a pulsed ultrasonic signal to a preset position on the container wall; the wavelength of the pulsed ultrasonic signal is less than the thickness of the container wall; The signal emission control module is used to control the emission process of the ultrasonic signal; The acquisition control module is used to control the acquisition process of the echo signal; The signal processing module is used to obtain liquid information data in the container through the analysis and processing of the ultrasonic echo signal; the analysis and processing are carried out through a machine learning model, that is, the characteristic data of the echo signal is input into the machine learning model, and the liquid information data representing the liquid information is obtained through the analysis and processing of the machine learning model; the liquid information data is compared with the constructed echo and liquid information database to obtain liquid characteristic information, and the liquid characteristic information includes liquid category information.

10. The liquid information detection system based on ultrasonic technology according to claim 9, wherein: The neural network adopts a BP neural network; the BP neural network includes an input layer, a hidden layer, and an output layer; the input layer is used to input the voltage amplitude of the echo signal; the output layer is used to output liquid information data, and the liquid information data includes liquid category information and / or liquid liquid level information; the BP neural network is trained using a standard data set with known liquid physical parameters, and the standard data set includes the echo amplitude of the liquid and the corresponding liquid types, and the data that has not been used for training is input into the trained BP neural network to detect different liquid types; the BP neural network is trained in the following manner: First, perform a single-echo training model, and the input layer of the BP neural network is used to input the single-echo amplitude A1 for training; Then, perform a double-echo training model, and the input layer of the BP neural network is used to input the single-echo amplitude A1 and the double-echo amplitude A2 for training; Then, perform a triple-echo training model, and the input layer of the BP neural network is used to input the triple-echo amplitudes A1, A2, and A3 for training; as the trained BP neural network; Finally, perform n - time echo training on the model. The input layer of the BP neural network is used to input the amplitudes A1, A2, A3 and A of n - time echoes n for training; and use it as the trained BP neural network; The BP neural network can be trained separately according to the amplitude A1 of the first echo, the amplitude A2 of the second echo, the amplitude A3 of the third echo, or the amplitude An of the nth echo n for separate training; The BP neural network is gradually trained through any combination of the amplitude A1 of the first echo, the amplitude A2 of the second echo, the amplitude A3 of the third echo, or the amplitude An of the nth echo. n The training model is gradually trained through any combination of the amplitude A1 of the first echo, the amplitude A2 of the second echo, the amplitude A3 of the third echo, or the amplitude An of the nth echo.

11. The liquid information detection system based on ultrasonic technology according to claim 9, characterized in that: The ultrasonic echo signal includes the amplitude of the echo signal and / or the waveform signal.

Images