Ultrasonic wave flight time online calibration design based on liquid neural network

Abstract

The invention relates to an ultrasonic wave flight time online calibration design based on a liquid neural network. The method comprises the following steps: selecting various factors influencing ultrasonic flight time to construct a data set for liquid neural network training; secondly, predicting the error between the ultrasonic flight time and the real flight time under specific influence factors through a liquid neural network model by utilizing the constructed data set, and finally, calling the established liquid neural network model to analyze the influence of each influence factor on the ultrasonic flight time error. The result shows that the method has the effects that the error value of the ultrasonic flight time and the real flight time under each influence factor can be quickly predicted on the basis of the original experimental data, and the influence degree of each influence factor on the error value of the ultrasonic flight time and the real flight time can be clearly displayed; the method is high in prediction precision and good in applicability and stability, and the calibration efficiency of the ultrasonic flowmeter can be improved.

Description

Technical Field

[0001] The invention relates to an ultrasonic flight time online calibration design based on liquid neural network, and belongs to the field of ultrasonic flowmeters. Background Art

[0002] Ultrasonic flowmeter is an instrument that measures fluid flow based on the ultrasonic principle. It is widely used in industries, environmental protection, petroleum and other fields. Its working principle is usually based on the time-of-flight method or the Doppler effect. Among them, the time-of-flight method is more common in liquid and gas flow measurement. This method uses two ultrasonic transducers located opposite each other in the fluid pipeline to calculate the flow velocity of the fluid by the difference in flight time when the ultrasonic wave propagates in the fluid along the same and opposite paths as the flow direction. The size of the flight time difference is directly proportional to the flow velocity of the fluid, so the flow rate can be determined.

[0003] With the rapid development of current industry and technology, ultrasonic measurement technology has been widely used in different fields, such as flow measurement, level monitoring and non-destructive testing. One of the core technologies of ultrasonic measurement is time of flight (TOF) measurement, and its accuracy directly affects the accuracy of the measurement results. However, due to the influence of external factors such as ambient temperature, pressure, and medium density, it is difficult for traditional TOF measurement methods to adjust measurement parameters in real time, resulting in certain errors. Especially when measuring in liquid media, the complexity of liquid properties and flow states further aggravates the difficulty of calibration, so it is particularly important to achieve real-time calibration of TOF measurement. Liquid Neural Networks (LNN), as an emerging neural network structure, has gradually attracted the attention of researchers because of its excellent performance in processing dynamic and time-varying data. The LNN structure has unique advantages in fluid measurement and dynamic signal processing, and its characteristics based on time series data are very suitable for solving the real-time calibration problems encountered in TOF measurement. Compared with traditional static neural networks, LNN can adapt to changes in the external environment in real time under limited training data, thereby dynamically adjusting model parameters and achieving higher-precision measurement calibration. Summary of the invention

[0004] The purpose of the present invention is to improve the accuracy and stability of ultrasonic time-of-flight (TOF) measurement in real time through online calibration technology based on liquid neural network. Through this method, adaptive and dynamic calibration can be achieved in complex liquid environments affected by temperature, pressure and medium characteristics, significantly reducing measurement errors, meeting the needs of high-precision and automated measurement, and helping to make industrial detection processes intelligent and efficient. The specific steps are as follows:

[0005] Step S1: According to the measurement principle of ultrasonic flowmeter time difference method, the key data of each experiment is screened, and the geometric factors, signal factors and flow field factors that affect the ultrasonic flight time are selected as the input parameters of the liquid neural network. These key factors include the transducer installation angle, pipe outer diameter, channel wall thickness, fluid flow rate, forward and reverse flow time, medium density and composition, fluid flow state and ambient temperature, so as to construct a data set and provide comprehensive and accurate data support for the training of the liquid neural network model.

[0006] Step S2: The geometric parameters, signal factors and flow field factors that affect the ultrasonic flight time measurement are used as the input of the liquid neural network, and the calibration error value of the ultrasonic flight time is used as the output of the liquid neural network to train the liquid neural network. After the training is completed, a liquid neural network model that has been built and can be directly called can be obtained.

[0007] Step S3: Call the liquid neural network model built in step S2. When it is known that some influencing factors among the geometric parameters, signal factors and flow field factors in the ultrasonic flight time measurement have a greater impact on the flight time measurement accuracy, the characteristic information of the corresponding parameter value is used as input, and the liquid neural network model is used to predict the influence of each parameter value on the ultrasonic flight time calibration accuracy.

[0008] Step S4: Call the liquid neural network model built in step S2 to analyze and quantify the specific impact of each influencing factor value on the ultrasonic flight time measurement, and evaluate the effect of each factor on the measurement accuracy.

[0009] Determining the input parameters in the prediction model in step S1 includes: 1) The transducer installation angle θ directly affects the propagation path length of the ultrasonic wave and the signal reception quality. Angle deviation from the optimal value will lead to a longer path, increase the flight time (TOF), and cause signal attenuation, thereby reducing measurement accuracy and resolution. 2) The outer diameter D of the pipe directly affects the propagation path length and flight time (TOF) of the ultrasonic wave. The larger the outer diameter, the longer the path and the increased TOF, which affects the measurement accuracy. In addition, the outer diameter will also change the flow field distribution, which will indirectly affect the flight time. Therefore, accurate calibration of the outer diameter of the pipe is crucial to ensure measurement accuracy. 3) The wall thickness of the sound channel has a significant impact on the propagation path and signal strength of the ultrasonic wave. The increase in wall thickness will lengthen the path of the ultrasonic wave in the pipe, resulting in a longer flight time (TOF), and will also cause signal attenuation, reducing the accuracy and stability of the measurement. 4) The fluid velocity v directly affects the ultrasonic flight time (TOF). When the flow rate increases, the TOF in the downstream direction will be shortened, and the TOF in the upstream direction will be prolonged, forming a time difference. This time difference is used to calculate the actual flow rate, so the fluid flow rate is a key influencing factor in TOF measurement. The higher the flow rate, the higher the requirements for measurement accuracy and response time. Accurate control and calibration of flow rate parameters are crucial to ensure measurement accuracy. 5) Downstream and upstream transit time t1, t2. According to the ultrasonic downstream transit time formula: Ultrasonic countercurrent transit time formula: The time of upstream and downstream flow will be affected by external factors, thus affecting the ultrasonic flight time. 6) The density and composition of the medium directly affect the propagation speed c of the ultrasonic wave. Denser media usually slow down the propagation speed of the ultrasonic wave, resulting in a longer flight time. Media of different compositions may have different sound speeds, resulting in deviations in the flight time. 7) The flow state of the fluid has a significant impact on the ultrasonic flight time. For laminar flow, the fluid velocity distribution is relatively uniform, and the ultrasonic propagation speed and flight time are relatively stable. In the turbulent state, the velocity and density in the fluid are unevenly distributed, causing the ultrasonic propagation path and speed to fluctuate, thereby causing fluctuations and errors in the flight time. 8) Temperature changes will directly affect the density and sound speed of the medium. Usually, when the temperature rises, the density of the liquid decreases and the sound speed increases, resulting in a shortened flight time of the ultrasonic wave; conversely, when the temperature decreases, the sound speed decreases and the flight time is extended. 9) Deviation of actual ultrasonic flight time. According to the flight time formula: The ultrasonic flight time is calibrated, and the ultrasonic flight time indication error of the calibration device is used as the actual flight time deviation for prediction, and the influence of various influencing factors on the ultrasonic flight time is analyzed.

[0010] The steps for constructing the liquid neural network in step S2 are: Step S2.1: Before starting to train the liquid neural network, initialize the parameters of the network, including the connection weights and dynamic response parameters of the neurons. The liquid neural network ensures the diversity of model parameters and promotes the randomness and stability of training by randomly initializing the values. Step S2.2: Randomly select an input sample and its corresponding expected output from the validation set. In the forward propagation stage, the input data is propagated through the dynamic liquid neuron layer in the network. Due to the time dependence of the liquid neural network, the input signal is gradually propagated to the output layer according to the dynamic state and activation function of each layer to preserve the timing information of the input data. Step S2.3: In the liquid neuron layer, the activation function performs a nonlinear transformation on the input data and adjusts the output according to the dynamic changes of the time series. By introducing dynamics, the liquid neural network can learn and recognize complex timing patterns and improve the accuracy of the prediction of ultrasonic flight time deviation. Step S2.4: The neurons in the output layer generate the final prediction results of the network. To ensure that the output meets the target requirements, the output layer of the liquid neural network will adjust the activation function according to the specific application. Step S2.5: Calculate the loss function to measure the performance of the model by comparing the error between the predicted value and the target value. The value of the loss function will provide feedback for the optimization of the model. Step S2.6: Enter the back propagation phase and calculate the gradient of each parameter to the loss function through the chain rule and gradient descent method. In this process, the liquid neural network combines the dynamic characteristics of the time series to effectively adjust the parameters and improve the learning effect of the model. Step S2.7: Based on the gradient information, use an optimization algorithm (such as stochastic gradient descent, momentum method, Adam, etc.) to update the network parameters. In multiple iterations, the network gradually reduces the loss value, thereby improving its ability to fit the time series data. Step S2.8: Forward propagation, loss calculation, and back propagation together constitute a training iteration of the liquid neural network. After multiple iterations, the network gradually optimizes the parameters, thereby enhancing the prediction ability of ultrasonic flight time deviation and adapting to complex dynamic data.

[0011] The ultrasonic flight time online calibration prediction step based on liquid neural network in step S3 is: Step S3.1: According to the selected environmental influencing factors (such as pipeline outer diameter, medium density and composition, transducer installation angle, fluid flow state, etc.), the experimental data are organized into the input and output format of the liquid neural network. The first 71% of the data is used as the training set, and the last 29% is used as the validation set to verify the accuracy. Step S3.2: Set the neural network parameters according to the number of inputs and outputs, and perform liquid neural network training. Step S3.3: After completing multiple training iterations and establishing a liquid neural network that meets the error requirements, input the transducer installation angle, pipe outer diameter, sound channel wall thickness, fluid flow rate, forward and reverse flow time, medium density and composition, fluid flow state, ambient temperature and the actual measured ultrasonic flight time deviation value Δt to predict the corresponding measured flight time deviation Δt1 value.

[0012] The specific steps of analyzing the influence of each influencing factor of the constructed liquid neural network on the ultrasonic flight time measurement accuracy in step S4 are: Step S4.1: Use the control variable method to fix the values ​​of other influencing factors and only adjust a certain influencing factor under study so that it changes gradually within a preset range with a fixed step size, and organize this series of changing data into the input form of the liquid neural network. Step S4.2: Call the liquid neural network model trained in step S2, take the optimal values ​​of key parameters in the liquid neural network to build the model, and finally input the validation set data into the model to obtain the predicted flight time deviation and the influence of various influencing factors on the ultrasonic flight time measurement accuracy. Step S4.3: When analyzing the influence of other influencing factors, select the corresponding influencing factors and repeat the above steps S4.1 and S4.2 until the influence analysis of each influencing factor on the ultrasonic flight time measurement accuracy is completed and the predicted ultrasonic flight time deviation result is obtained, thereby realizing online calibration.

[0013] The beneficial effects of the present invention are as follows: as long as there are experimental data from the test, a liquid neural network can be quickly built on the basis of the experimental data to realize the prediction of the ultrasonic flight time deviation under any influencing factors, and the influence of each influencing factor on the ultrasonic flight time measurement accuracy can be accurately analyzed, thereby realizing online calibration, the process is stable and reliable, and the efficiency is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is a flow chart of the design prediction and influencing factor analysis steps of ultrasonic flight time online calibration based on liquid neural network described in the patent of this invention;

[0015] Figure 2 It is the flow chart of the liquid neural network algorithm described in the patent of this invention;

[0016] Figure 3 This is a flow chart of the design steps for online calibration of ultrasonic flight time based on liquid neural network described in the patent of this invention. DETAILED DESCRIPTION

[0017] The method of the present invention is further described in detail below with reference to the embodiments and accompanying drawings:

[0018] like Figure 1 As shown, the ultrasonic flight time online calibration design based on liquid neural network described in the present invention includes the following specific implementation steps:

[0019] Step S1, selecting factors that affect the measurement accuracy of ultrasonic flight time: Under the premise of the time difference method principle, ultrasonic flight time is closely related to the transducer installation angle, pipe outer diameter, sound channel wall thickness, fluid flow rate, upstream and downstream time, medium density and composition, fluid flow state and ambient temperature.

[0020] Step S2, training and building the liquid neural network according to the constructed data set: Figure 2 As shown in the figure, a prediction model for ultrasonic flight time online calibration deviation based on liquid neural network is established, which specifically includes the following steps: 1) According to the experimental data, the transducer installation angle, pipe outer diameter, sound channel wall thickness, fluid flow rate, forward and reverse flow time, medium density and composition, fluid flow state, ambient temperature and the deviation value of the actual ultrasonic flight time are input to predict the flight time deviation. The first 71% of the data is used as the training set, and the last 29% is used as the validation set to verify the prediction accuracy of the algorithm. 2) Set the parameters of the liquid neural network, such as the number of layers, number of nodes, and activation function, and perform multiple rounds of training based on the input features and output deviation values. The liquid neural network continuously optimizes the model parameters through back propagation and error adjustment, gradually reducing the prediction error. 3) After completing multiple training iterations, verify the prediction accuracy of the liquid neural network model, use the trained liquid neural network for ultrasonic flight time deviation prediction, and output the predicted value of flight time deviation under each input condition. Ultimately, the liquid neural network model can stably predict the deviation of ultrasonic flight time due to various influencing factors, improving measurement accuracy and calibration efficiency.

[0021] Step S3, calling the constructed liquid neural network to predict the ultrasonic flight time deviation under a certain influencing factor; Figure 3 As shown, by calling the established liquid neural network and inputting the corresponding influencing factor value, the ultrasonic flight time deviation can be predicted. In this process, the last 29% of the experimental data is input into the established liquid neural network as a validation set, and the predicted flight time deviation is compared with the actual flight time deviation to verify the prediction accuracy of the model.

[0022] Step S4, calling the constructed liquid neural network model to analyze the influence of various influencing factors on the measurement accuracy of ultrasonic flight time: when studying the influence of geometric parameters, signal factors and flow field factors on ultrasonic flight time, by controlling the value of a single variable, other factors are gradually increased with a fixed step size within the interval, and this set of data is sorted into the input format of the liquid neural network, and input into the constructed liquid neural network model, the deviation caused by the change of ultrasonic flight time with other factors can be obtained. By analyzing the influence of various influencing factors on ultrasonic flight time, the measurement accuracy of ultrasonic flight time can be effectively improved.

[0023] The method of the present invention is based on a liquid neural network to perform an online calibration of ultrasonic flight time, which mainly predicts the deviation of ultrasonic flight time to evaluate the measurement accuracy and stability of the ultrasonic flowmeter during actual use. The method can identify and monitor various factors affecting the performance of the equipment in real time, and adjust the measurement results in time through online error compensation technology, thereby achieving efficient online calibration. This process not only effectively extends the calibration cycle of the ultrasonic flowmeter and improves the efficiency of the instrument, but also avoids the potential impact of the traditional inspection process on the measurement results, ensuring the high accuracy and reliability of the ultrasonic flowmeter in practical applications.

[0024] The above examples are only used to help understand the core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A design of ultrasonic time-of-flight online calibration based on liquid neural network, including: Step S1 performs feature extraction on geometric parameters, signal factors and flow field factors that affect ultrasonic flight time and forms a data set for liquid neural network training; step S2 performs training and model building on the liquid neural network based on the processed data set; step S3 calls the built liquid neural network to predict the flight time deviation under a certain influencing factor among the geometric parameters, signal factors and flow field factors that affect ultrasonic flight time; and step S4 analyzes the ultrasonic flight time deviation under each influencing factor.

2. The factors affecting the ultrasonic flight time in step S1 according to claim 1 can be obtained from the formula It can be seen that geometric parameters, signal factors and flow field factors will affect the ultrasonic flight time.

3. According to claim 1, step S2 is to build a model of a liquid neural network for the data set, characterized in that Liquid neural networks use dynamic neurons based on differential equations, which can continuously update their states over time; neurons adapt to changes in external inputs by adjusting weights and states in real time.

4. According to step S3 of claim 1, the constructed liquid neural network model is used to predict the ultrasonic flight time deviation under a certain influencing factor, characterized in that The liquid neural network model uses historical and current input data to dynamically adjust the prediction results to cope with the complex impacts of flight time deviation in the actual environment.

5. According to step S4 of claim 1, the constructed liquid neural network model is called to analyze the ultrasonic flight time deviation under each influencing factor, characterized in that Through the dynamic adaptive ability and recursive feedback mechanism of the liquid neural network model, the specific influence of various influencing factors (such as temperature, flow rate, medium density and composition, etc.) on the flight time deviation is gradually analyzed. The liquid neural network can update the state of each neuron in real time based on the input influencing factors, capture the nonlinear characteristics of the influencing factors and their cumulative effects over time, thereby providing more accurate deviation analysis.

Images