A robotically intelligent tow tank for testing vortex-induced vibration and a method of experimentation thereof

Abstract

This invention belongs to the field of vortex-induced vibration experimental technology and discloses a robotic intelligent towed water tank for testing vortex-induced vibration. The tank includes a water tank, a towed robot, and a computer controlling the towed robot. The water tank includes a main carriage mounted on its top. The towed robot includes a three-degree-of-freedom platform mounted on the main carriage. The three-degree-of-freedom platform is connected to a cylinder with sensors located in the water tank. The sensors are electrically connected to the computer. The computer controls the three-degree-of-freedom platform to drive the cylinder in any trajectory or any combination of trajectories of coaxial, transverse, and rotational motion in the water tank. This invention can automatically perform a series of forced vibration experiments to study vortex-induced vibration. By minimizing a suitable function to select the parameters for the next experiment, this method greatly reduces the experimental burden and human intervention, automating the experimental process and improving efficiency.

Description

Technical Field

[0001] This invention belongs to the field of vortex-induced vibration experimental technology, specifically a robotic intelligent towing pool for testing vortex-induced vibration and its experimental method. Background Technology

[0002] The field of fluid-structure interaction is rich in physical complexity regarding fluid dynamics and aerodynamic flows. One typical problem in this field is vortex-induced vibration (VIV). VIV occurs when a flexibly mounted bluff body is placed in an oncoming crossflow. At Reynolds numbers (hereinafter referred to as "Re") of approximately Re = 50 and above, spontaneous instabilities in the wake of the object lead to the formation of asymmetric vortices, resulting in unstable loads on the object and thus a vibration response. As a function of Re, extensive forced vibration testing is required. Towed water tanks are a commonly used testing equipment in this field, but traditional towed water tanks have some technical drawbacks. In this experiment, the discovery that simulant motion significantly affects crossflow vibration further increases the complexity of the parameters. The number of independent parameters required to predict the vibration response of flexible structures in shear flow is enormous. Assuming we have a 10-dimensional parameter space, blindly performing 10 measurements for each parameter would require 10 billion experiments, which is clearly infeasible. Therefore, we need to improve existing towed water tank experimental equipment, endowing it with intelligence to automatically conduct a series of forced vibration experiments to study vortex-induced vibration. Summary of the Invention

[0003] The purpose of this invention is to provide a robotic intelligent towing pool for testing vortex-induced vibration and its experimental method, so as to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a robotic intelligent towing pool for testing eddy-induced vibration and its experimental method, comprising a pool, a towing robot, and a computer controlling the towing robot. The pool includes a main carriage mounted on its top. The towing robot includes a three-degree-of-freedom platform mounted on the main carriage. The three-degree-of-freedom platform is connected to a cylinder with sensors located in the pool. The sensors are electrically connected to the computer. The computer controls the three-degree-of-freedom platform to drive the cylinder to move along any trajectory or any combination of trajectories of coaxial, transverse, and rotational motion in the pool.

[0005] Preferably, it includes the following steps:

[0006] S1. Determine the input parameters and their range that can affect the target's quantity of interest and transmit this information to the computer;

[0007] S2; Subsequently, the parameter space is explored and utilized adaptively and sequentially, allowing the computer to control the towing robot to automatically execute experiments and record experimental data through sensors;

[0008] S3. Before a new experiment begins, the computer uses Gaussian regression to learn from the existing data and update the prediction of the target's interest quantity.

[0009] S4. Learning stops when the predicted target interest converges.

[0010] Preferably, in S2, the towed robot needs to be forcibly stopped for a period of time before it runs the next experiment to allow the fluid motion to calm down.

[0011] Preferably, the computer includes a storage medium containing a control program and an execution program, the execution program comprising the methods in S2-S4 of claim 2.

[0012] Preferably, the Gaussian regression method includes the following steps:

[0013] S3.1. Data Collection: It is necessary to collect input variables and corresponding output variables from the environment;

[0014] S3.2. Data Preprocessing: The collected data needs to be preprocessed, including data cleaning, noise reduction, normalization and other operations, in order to facilitate the training of the Gaussian regression model;

[0015] S3.3. Gaussian Regression Model Training: The collected data needs to be used to train the Gaussian regression model;

[0016] S3.4. Model Evaluation: The trained Gaussian regression model needs to be evaluated to ensure its accuracy and reliability.

[0017] S3.5. Autonomous Learning: Predicting output variables based on current input variables, and making decisions and adjustments based on the prediction results;

[0018] S3.6. Model Update: It is necessary to continuously collect data from the environment and use this data to update the Gaussian regression model in order to improve the accuracy and reliability of the model.

[0019] The beneficial effects of this invention are as follows:

[0020] This invention introduces active learning to give intelligent control to experimental devices such as towed pools, enabling them to automatically conduct a series of forced vibration experiments to study vortex-induced vibration. By minimizing a suitable function, the next experimental parameters are selected, which greatly reduces the experimental workload and human intervention. Automation accelerates the experimental process and improves efficiency. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the operation of the device of the present invention;

[0022] Figure 2 The diagram shows the sequence and typical results of the GPR learning process in this invention.

[0023] Figure 3 This is a schematic diagram illustrating the exploration of large parameter space for the present invention;

[0024] Figure 4 This is a schematic diagram of the three-degree-of-freedom platform structure of the present invention.

[0025] In the diagram: 1. A three-degree-of-freedom platform; 2. A water tank. Detailed Implementation

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] like Figures 1 to 4 As shown, this embodiment of the invention provides a robotic intelligent towing pool for testing eddy-induced vibration and its experimental method, including a pool, a towing robot, and a computer controlling the towing robot. The pool includes a main carriage mounted on its top, and the towing robot includes a three-degree-of-freedom platform mounted on the main carriage. The three-degree-of-freedom platform is connected to a cylinder with sensors located in the pool. The sensors are electrically connected to the computer. The computer controls the three-degree-of-freedom platform to drive the cylinder to move along any trajectory or any combination of trajectories of coaxial, transverse, and rotational motion in the pool.

[0028] A three-degree-of-freedom platform can be driven to perform combined in-line, transverse, and rotational motions, such as... Figure 4 As shown, the longitudinal and transverse guide rail mechanisms achieve coaxiality and constant current, while the cylinder with a sensor on the transverse guide rail can achieve rotational motion. The Experimental Technique (ITT) computer records motion and force signals. The ITT process begins with hypothesizing (this hypothesis is generated by humans, or may be synthesized collaboratively by robots, computers, and humans in the future). Then, the ITT performs adaptive sequential experiments to learn the target QoI, interrupted only by pauses between experiments to avoid cross-contamination of results between consecutive experiments. Upon convergence, the learned QoI is further post-processed to check the validity of the hypothesis. No human intervention is required during the continuous experimental testing.

[0029] An experimental method for testing vortex-induced vibrations in a robotic intelligent towing pool includes the following steps:

[0030] S1. Determine the input parameters and their range that can affect the target's quantity of interest and transmit this information to the computer;

[0031] S2; Subsequently, the parameter space is explored and utilized adaptively and sequentially, allowing the computer to control the towing robot to automatically execute experiments and record experimental data through sensors;

[0032] S3. Before a new experiment begins, the computer uses Gaussian regression to learn from the existing data and update the prediction of the target's interest quantity.

[0033] S4. Learning stops when the predicted target interest converges.

[0034] Leveraging subject-specific knowledge, we first identify the input parameters and their ranges that may influence QoIs and pass this information to ITT. Next, we can begin adaptively and sequentially exploring and utilizing the parameter space, automatically executing corresponding experiments to predict QoIs. After a new experiment begins, ITT uses Gaussian Regression (GPR) to learn from existing data to update its predictions of QoIs. Simultaneously, we find the input for the next experiment by minimizing the acquisition function, which describes the uncertainty as a function of parameters, specifically the standard deviation (SD). A forced pause is performed before running the next experiment to allow the fluid motion to settle and to avoid cross-contamination of results. Then, after collecting new data, the next iteration of the learning process begins. Learning stops when the prediction of QoIs converges. Here, we track the maximum value of SDσmax during iterations to be consistently less than a reference level. Because GPR learning is a stochastic process, this convergent reference should be associated with inherent system uncertainties. Due to modeling and measurement, we categorize system uncertainties into two types. Modeling uncertainty stems from the choice of alternative models and optimization methods used for learning, while measurement uncertainty arises from sensor noise and problem-related physical uncertainties, such as background turbulence in the water. Ideally, if we can perfectly map the unknown function of QoI, there will be zero modeling error, and therefore, the uncertainty of prediction will converge to the uncertainty of measurement. Measurement uncertainty is an inherent property of the experimental setup (varying from setup to setup) and must be calibrated beforehand.

[0035] As the number of experiments increased, ITT revealed more details about the different parameters (Clv, fr, and Ay / d). Meanwhile, in Figure 2 In each iteration of B, the σmax value was found to decrease as it approached the 3σr reference line. (This was observed in 36 experiments.) Figure 2 A.3 and 37 experiments Figure 2 Between A.4, a new feature appeared in the second positive region of Clv, accompanied by... Figure 2 A slight increase in σmax is shown in B. Finally, upon convergence, ITT has learned the Clv mode.

[0036] In step S2, the towed robot needs to be forcibly stopped for a period of time before it can calm the fluid motion.

[0037] The computer includes a storage medium containing a control program and an execution program, wherein the execution program includes the methods in S2-S4 of claim 2.

[0038] The specific practical methods and processes of this invention are as follows:

[0039] Using a newly constructed ITT, we investigated one of the typical fluid-structure interaction problems—the VIV of a bluff body—through a forced vibration experiment on a rigid cylinder. This study realizes an idea that is not new to researchers in scientific robotics. It demonstrates that by carefully calibrating the inherent uncertainties of the experimental apparatus and selecting appropriate machine learning tools (in this study, the foundation and core functions of GPR for QoIs), the ITT can adaptively and intelligently design and conduct sequential experiments to study the target QoIs (in this study, the hydrodynamic coefficients of a cylinder forced to vibrate in a uniform flow); the complex physics of nonlinear systems with the same level of accuracy but orders of magnitude less compared to traditional experimental sampling strategies; and the exploration of a wider parameter space (up to eight parameters in this study) for new physical insights and scaling, which was not feasible in past studies, thus accelerating scientific discovery.

[0040] The most important method used in this device is ② Gaussian Process Regression (GPR). GPR is a non-parametric method that models and predicts unknown functions from a finite set of training points. It has been successfully applied in various fields to explore hidden physical phenomena in data. For example, in this device, the robot conducts ten experiments to measure and adjust the velocity, amplitude, and frequency of a cylinder in a pool. GPR predicts the optimal solution for the eleventh experiment based on this data. The most important aspect of the GPR learning result is that when the learning process stops, the ITT (Integrated Technology for Testing and Research) not only provides a set of experimental data, but more importantly, it provides an accurate functional approximation of the target QoI. When applying the obtained data to predict or understand more complex problems, this functional representation opens up new possibilities for using various optimization tools, while incorporating additional physical insights as constraints.

[0041] ③ The Morris method for global sensitivity analysis is used to collect input parameters, sample the parameter space, and evaluate the output of each model to screen important input parameters for robot models or problems.

[0042] We describe how our research, in parallel with work in other fields, reduces by orders of magnitude the number of experiments required to explore and map the complex fluid dynamic mechanisms controlling fluid elastic instability and the resulting nonlinear VIV response. We demonstrate the effectiveness of the “exploration-exploitation” approach in high-dimensional parameter spaces, which is difficult to address using traditional methods that vary system parameters in experiments.

[0043] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0044] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A robotic intelligent towed pool for testing vortex-induced vibrations, comprising a pool, a towed robot, and a computer for controlling the towed robot, characterized in that: The pool includes a main slide mounted on its top, and the towing robot includes a three-degree-of-freedom platform mounted on the main slide. The three-degree-of-freedom platform is connected to a cylinder with sensors located in the pool. The sensors are connected to a computer for data transmission. The computer controls the three-degree-of-freedom platform to drive the cylinder to move along any trajectory or any combination of trajectories of coaxial, transverse, and rotational motion in the pool. The computer is configured to control the towed robot to automatically perform the following experimental steps: S1. Determine the input parameters and their range that can affect the target's quantity of interest and transmit this information to the computer; S2. Subsequently, the parameter space is explored and utilized adaptively and sequentially, allowing the computer to control the towing robot to automatically execute experiments and record experimental data through sensors; S3. Before a new experiment begins, the computer learns from the existing data using Gaussian regression, updates the prediction of the target interest quantity, and finds the input for the next experiment by minimizing the acquisition function, which describes the uncertainty through the standard deviation of the function as a parameter. S4. Learning stops when the predicted target interest converges; convergence means that the maximum value of the standard deviation during the tracking iteration is stably less than the reference level, which is associated with the inherent measurement uncertainty, which comes from sensor noise and background turbulence in the water. The measurement uncertainty is an inherent property of the experimental equipment and has been calibrated in advance. When the predicted uncertainty converges to the measurement uncertainty, it represents zero modeling error. In S2, the towed robot needs to be forcibly stopped for a period of time before it runs the next experiment to allow the fluid movement to calm down. The Gaussian regression method includes the following steps: S3.

1. Data Collection: This requires collecting input variables and corresponding output variables from the environment; S3.

2. Data Preprocessing: The collected data needs to be preprocessed, including data cleaning, noise reduction, and normalization, to facilitate the training of the Gaussian regression model; S3.

3. Gaussian Regression Model Training: The collected data needs to be used to train the Gaussian regression model; S3.

4. Model Evaluation: The trained Gaussian regression model needs to be evaluated to ensure its accuracy and reliability; S3.

5. Autonomous Learning: Predicting output variables based on current input variables, and making decisions and adjustments based on the prediction results; S3.

6. Model Update: It is necessary to continuously collect data from the environment and use this data to update the Gaussian regression model in order to improve the accuracy and reliability of the model.

Images