In this paper, we present a model-free deep reinforcement learning (DRL) algorithm to control the locomotion of a quadruped robot adaptively at challenging terrain. This approach is used to improve the robot’s abilit...
In this paper, we present a model-free deep reinforcement learning (DRL) algorithm to control the locomotion of a quadruped robot adaptively at challenging terrain. This approach is used to improve the robot’s ability to traverse unstructured ground. Combining Bidirectional Gate Recurrent Unit (BGRU) and attention mechanism, we train the quadruped robot to climb up and down stairs, traverse various field terrain and bricks. We train the controller in simulation using knowledge distillation, and deploy on real robots without any adjustment. In the simulation, the performance of our Student Module even exceeds that of Teacher Module. The experimental results demonstrate that our proposed approach can make the quadruped robot traverse obstacles swiftly.
A dynamic image reconstruction method considering the spatiotemporal evolution characteristics of time-varying distribution is proposed for electrical resistance tomography (ERT). The dynamic inversion problem of ERT ...
A dynamic image reconstruction method considering the spatiotemporal evolution characteristics of time-varying distribution is proposed for electrical resistance tomography (ERT). The dynamic inversion problem of ERT is constructed by state-space modeling method with state evolution and observation update equations, and is solved by Kalman filter. To accurately describe the state evolution process of time-vary parameters, the latent variable based statistical modeling method is proposed to construct the state evolution equation. The potential characteristics of the state parameters in the dynamic change process are fully explored and characterized from the data with multivariate regression methods. Numerical and experimental results show that the proposed dynamic image reconstruction method can improve the imaging quality of ERT for time-varying distribution.
The process states in gas-liquid two phase flow are the synthetical consequence of multiple apparent factors related to time, space, flow conditions and so on. Higher-order tensor decomposition in multilinear algebra ...
The process states in gas-liquid two phase flow are the synthetical consequence of multiple apparent factors related to time, space, flow conditions and so on. Higher-order tensor decomposition in multilinear algebra provides a powerful framework mathematically to analyze the multifactor structure of flow state ensembles. Focusing on entangling the constitute factors or modes embedded in various states, a novel strategy about multilinear subspace analysis of state ensembles in gas-liquid two-phase flow via pulse wave ultrasonic Doppler (PWUD) signal is proposed. PWUD sensor is adopted to acquire spatiotemporal information of flow states under varying flow conditions. Based on processed PWUD signal, higher order singular value decomposition (HOSVD) is employed for tensor analysis of flow state ensembles that combine several modes, including different state categories, time, space and flow conditions. The purpose of the strategy is to perform few-shot learning on the foundation of refining apparent factors underlying the formation of flow states with a simple sensor, that is, using limited state samples as reference to further infer the states under different flow conditions. The experimental results demonstrate that multilinear representation yields improved state identification rates relative to the conventional single-factor method principal component analysis (PCA).
Pressure ulcers are a particularly high incidence of chronic trauma. When a superficial wound is visible, the underlying wound is often serious, so early detection of pressure ulcers is critical. Currently there is st...
Pressure ulcers are a particularly high incidence of chronic trauma. When a superficial wound is visible, the underlying wound is often serious, so early detection of pressure ulcers is critical. Currently there is still no technique or real-time monitoring system that can visualize progressive tissue damage of pressure ulcers. Electrical impedance tomography (EIT) is a functional imaging technique that can diagnose the health of the tissue by visualizing the distribution of bioelectrical impedance parameters. Based on the differences between the electrical properties of pressure ulcers and normal tissue, a non-invasive pressure ulcer depth detection method based on EIT is proposed. Given surface voltage data measured on an open rectangular electrode array, the conductivity distribution under the skin surface was reconstructed to obtain pressure ulcer depth information which is expected to visualize progressive tissue damage. Based on EIT combined with flexible sensor arrays, finite element simulation models and tissue-mimicking agar phantom models were established to verify the effectiveness of the EIT-based depth detection method of pressure ulcers. The effects of time-difference imaging and frequency-difference imaging on the depth detection of pressure ulcer were compared. Both simulation and physical phantom experiments demonstrated the feasibility of EIT for detection of early pressure ulcer depth.
The intermittent slug flow in horizontal circular pipes occurs commonly in industrial applications such as pipeline transportation and nuclear reactor cooling. The phenomenon of elongated bubble centring in slug flow ...
The intermittent slug flow in horizontal circular pipes occurs commonly in industrial applications such as pipeline transportation and nuclear reactor cooling. The phenomenon of elongated bubble centring in slug flow can affect the bubble interface morphology, resulting in fluctuations of characteristic parameters. Therefore, it is important to acknowledge the extent to which bubble centring occurs and the possible influencing factors. This paper aims to investigate the elongated bubble centring in moderately sized pipes based on the laser-induced fluorescence method. A new image processing procedure is developed to measure the centring degree. A novel segmentation method is proposed to improve measurement accuracy by combining the global threshold algorithm with the Grey Wolf Optimization. The statistical analysis shows a positive correlation between the Froude number and the degree of bubble centring. Furthermore, the effect of superficial flow rate ratios is explored in the cases of similar Froude numbers. Finally, the results are verified by the models and the experimental data previously published.
Gas-water two-phase flow is a complex and changeable process, which widely exists in a variety of industrial scenarios. The gas-water two-phase flow has complex flow characteristics and fluid structure, which can be c...
Gas-water two-phase flow is a complex and changeable process, which widely exists in a variety of industrial scenarios. The gas-water two-phase flow has complex flow characteristics and fluid structure, which can be changed into different flow statuses. There are both differences and similarities among various flow statuses. In order to realize the accurate identification of gas-water two-phase flow, a jointly specific and shared dictionary learning (JSSDL) method is proposed. Based on multi-sensor data, the specific discrimination dictionary for different flow statuses and the shared dictionary for similar characteristics are established respectively to solve the problem of gas-water two-phase flow status identification. In the horizontal gas-water two-phase flow circuit, the effectiveness of the method is verified by experiments.
Gas-liquid two-phase flow is a common flow phenomenon in industrial processes, with typical characteristics of complex processes such as complex structure, large fluctuation and various variables. In order to achieve ...
Gas-liquid two-phase flow is a common flow phenomenon in industrial processes, with typical characteristics of complex processes such as complex structure, large fluctuation and various variables. In order to achieve accurate monitoring of the gas-liquid two-phase flow status, a monitoring method based on hidden Markov model (HMM) is proposed, which can analyze the dynamic characteristics of the flow with good performance. HMM contains continuously changing hidden states, which is consistent with the structure alternation and strong fluctuation that characterize gas-liquid two-phase flow. The principal component analysis (PCA) method is used to extract features, and the Gaussian mixture model (GMM) is used to fit the data. Since the modeling effectiveness of the HMM is significantly affected by its stochastic initial parameters, the genetic algorithm (GA) is used to reduce the sensitivity of the initial parameters. Continuous hidden Markov models are established and the current flow status can be monitored online according to the probabilistic indicators. The applications to monitoring five typical flow statuses and transition status from plug flow to slug flow demonstrate the validity and accuracy.
As a kind of flow process with complex and variable flow structure, multiple process variables and nonlinear characteristics, the flow status of multiphase flow shows non-stationary characteristics, and it is difficul...
As a kind of flow process with complex and variable flow structure, multiple process variables and nonlinear characteristics, the flow status of multiphase flow shows non-stationary characteristics, and it is difficult to achieve accurate online monitoring. For the nonlinear characteristics of the flow status of gas-liquid two-phase flow, a monitoring method based on principal polynomial analysis is proposed to extract the nonlinear characteristics of the flow status information. This method performs dimensionality reduction in multiple steps and uses polynomials for feature fitting, which can fully extract flow status information. Multiple sensors are used to obtain data representing different fluctuation characteristics in the flow process, models of different flow statuses are established, and status monitoring is realized based on statistical indicators. In the comparison with the principal component analysis method, it can be found that the principal polynomial analysis method can extract the nonlinear features and can describe the flow status more comprehensively, which can improve the accuracy of status monitoring.
Droplet size distribution is a key parameter to predict the pressure drop and heat transfer in annular flow. A novel fiber optical reflectometer technique is used to measure the droplet size of annular flow in a DN50 ...
Droplet size distribution is a key parameter to predict the pressure drop and heat transfer in annular flow. A novel fiber optical reflectometer technique is used to measure the droplet size of annular flow in a DN50 horizontal pipe. The experiment is carried out in the superficial gas velocity range from 15 m/s to 18 m/s and superficial liquid velocity range from 0.04 m/s to 0.35 m/s. The results reveal that droplet size is strongly dependent on the gas and liquid flow velocities. When the superficial liquid velocity increases from 0.04 m/s to 0.35 m/s, the droplet Sauter diameter increases from 190 μm. to 337 μm. on average. Then, based on the available experiment data in the previous literature, five published models of droplet size are summarized and evaluated, and the differences among these models are analyzed. Finally, a new empirical model is established for the prediction of droplet size, which takes into account the effects of Reynolds number, liquid viscosity, pipe diameter, gas-liquid density ratio and system pressure. Compared with the published correlations, the new model improves the prediction accuracy of droplet size. The average relative error between experimental data and prediction values is significantly reduced to 10.71%, and 96% of the predicted results have a relative error within 25%, which verifies the predictability of the new droplet size model in a wide range of experimental conditions.
Ultrasonic flowmeter has been widely used in chemical industry, metallurgy, and oil industry due to the advantage of highly accurate, non-invasive, and no pressure loss nature. However, in some harsh measurement scena...
Ultrasonic flowmeter has been widely used in chemical industry, metallurgy, and oil industry due to the advantage of highly accurate, non-invasive, and no pressure loss nature. However, in some harsh measurement scenarios like in hydrogen recycling loop and downhole energy fluid transportation, it may lack the accuracy to resolve small flow rate change. Traditional Time-Of-Flight (TOF) estimation methods used in ultrasonic flowmeters tend to show low accuracy and lack of robustness in these scenarios which often tend to show low signal-noise ratio and with a lot of noise interference. So, the key factor is the TOF estimation accuracy. To solve this problem, we propose the central frequency matching method. With the use of segmentation, signal filtering, and time-frequency analysis, it can provide TOF estimation accuracy and noise robustness. According to the experimental study, the proposed method makes good use of the characteristics of ultrasonic flow meters, improves the accuracy and reduces the error, and effectively broadens the applicable scenarios of ultrasonic flow meters.
暂无评论