Aero flywheel motors integrate a starter, flywheel, and generator in a heavy fuel piston aero-engine. Direct torque control (DTC) gives quick dynamic response, has a simple structure, and is very robust so it is the best choice for controlling aero flywheel motors. The flux estimator, the most important part of the controller, is influenced by the initial value, DC offset, resistor error, and inverter nonlinearity. This paper describes a flux estimator based on a low pass filter (LPF) to improve the estimates. Simulations of the aero flywheel motor starting performance with this flux estimator show that that the starting performance is improved.

According to requirement of the high-temperature application from the digital image correlation (DIC) method, the speckle fabrication methods and high-temperature speckle image processing technique are studied in this paper. Taking high-temperature structural materials as the studied objects, a high-temperature speckle fabrication technique based on parametric template is developed. The technique is mainly based on a parametric template, which can adjust the parameters such as particle size, distribution density, and randomness in the template. As a result, the quality of the speckle can be optimized. A high-temperature speckle pattern was successful fabricated on the C-SiC composite substrate by the proposed technique, and the experiment verifies the speckle carrier can endure high temperature up to 1 200℃. At the same time, the influence factors of thermal disturbance and thermal radiation on DIC measurement results are analyzed. The high-temperature speckle images are corrected with approaches of image grayscale-average, circulating thermal disturbance and optical filtering method. And a test for measuring the thermal expansion coefficient is designed to verify the feasibility of the correction methods.

The burning characteristics of aircraft cabin materials must be well characterized to enable prevention and early detection of in-flight fires. This study characterized two aircraft cabin materials for the seat and the interior wall at Kangding Airport in Sichuan Province where the atmospheric pressure is 61.0 kPa and at Guanghan where the pressure is 96.0 kPa. The experiments measured the mass loss ratio, smoke density, smoke components, and flame shape to examine the effect of low pressure on the burning characteristics. The results show that both materials have higher mass loss ratios at 96.0 kPa than at 61.0 kPa. However, the smoke density increases rapidly and changes more at 61.0 kPa. The smoke density peaks for both materials at 61.0 kPa are twice those at 96.0 kPa. Although the times to reach the peak concentrations of CO and CO₂ are nearly the same, the concentrations of CO and CO₂ are quite different for the two pressures.

Robotic drilling and riveting systems use visualization to precisely position the workpiece and for quality assurance. This paper describes a multi-step method to improve the visual detection accuracy of the reference and connection holes. A pixel-level edge point set based on the Canny operator is used first for coarse positioning of the circular hole. The edge parameters of the three-grey transition model are accurately derived using four Zernike orthogonal moments for the region of interest around the hole to be detected. Then, the relationships between the edge parameters and thresholds is used to calculate the subpixel coordinates of the edge points. A least squares analysis is then used to detect the center location and hole radius. Simulations show that the error in the center coordinates is approximately 0.01 pixels and that of the radius is 0.1 pixels, which is higher detection accuracy than the traditional algorithm. The system works for various size holes and is less sensitive to noise. Thus, this method effectively improves the detection precision of circular holes, which is important in robotic drilling and riveting visual measurement systems.

Visual sensors based on laser structured light are widely used in the welding field for detection of welding grooves and tracking of weld seams. This paper describes a visual sensor based on combined laser structured light where the light paths within the visual sensor are designed to eliminate the need to calibrate the external sensor parameters for different positions and postures to provide on-line detection of welding groove parameters using only the internal sensor parameters (which require calibration) and the feature point coordinates of the welding groove image. Thus, this method is much more adaptable than previous methods. The image processing is optimized by improving and combining various image processing algorithms. A dynamic region of interest region extraction algorithm is used to quickly find the region that includes the laser lines and feature points to significantly improve the subsequent image processing speed. Then, a top hat transformation and self-adaptive binarization are used to homogenize the gray values of the laser lines that differentiates the laser lines from the background. After that, an edge recognition algorithm based on the Laplacian of Gaussian (LOG) operator is used to extract the single pixel edges of the laser lines which are used to then identify the discrete points of the laser center lines that are fit to linear equations by the least squares method. Finally, these linear equations are used to accurately calculate the image coordinates of the feature points of the welding groove. All these image processing algorithms were implemented in Visual Studio for tests to detect V-shaped welding grooves with various characteristic parameters. The detection errors are all within ±4%, which shows that the detection algorithms are reasonable and applicable for image processing of welding processes with visual sensors.

The vertical ground reaction force determines the movement of the center of mass of the human body in the vertical direction during stair climbing. The vertical reaction force is closely related to the average vertical speed and the initial vertical speed. This paper presents a human stair climbing model that is used to derive the constraint for continuous climbing from the key factors influencing the maximum vertical force. The model gives mathematical expressions for the initial vertical speed, vertical force and the average horizontal speed. The model gives the change in the maximum vertical ground reaction force for various stair climbing speeds. Stair climbing tests were performed at various speeds using a motion analysis system and force platform. The tests verify the model accuracy with the results showing that the motion characteristics at low stair climbing speeds are closer to those of level walking than at high speeds and the initial vertical speed increases with increasing average horizontal speed to reduce the maximum vertical ground reaction force.

Named entity recognition (NER) in Chinese social media is less effective than in standard news mainly due to the normalization and the size of the existing annotated corpus. In recent years, research on named entity recognition in Chinese social media has tended to use external knowledge and joint training to improve performance due to the small size of the annotated corpus. However, there are few studies on mining entity recognition characteristics in social media. This article focuses on named entity recognition in text articles using a neural network model that combines bi-directional long short-term memory with a self-attention mechanism. This model extracts context information from different dimensions to better understand and represent the sentence structure and improve the recognition performance. Tests on the Weibo NER released corpus show that this method is more effective than previous approaches and that this method has a 58.76% F₁-score without using external knowledge or joint learning.

Traditional eye movement models are based on psychological assumptions and empirical data; thus, they cannot predict eye movement for previously unseen text and cannot predict individual differences while reading. This paper presents an eye movement model based on conventional psychology-based eye movement models using a bi-directional long short-term memory-conditional random field (bi-LSTM-CRF) neural network instead of empirical data sets. The model was trained to predict the eye movements of a user reading a previously unseen text based on the eye movements recorded for this person reading other texts as training data. Tests demonstrate that the model can achieve similar prediction accuracy than current machine learning models while requiring fewer features, which makes this model attractive for a range of real-time human-computer applications.

A Gaussian mixture model-hidden Markov model (GMM-HMM) for speech recognition uses the most likely state sequence (MLSS) criterion to get the best state series of observations. Since the MLSS search algorithm only considers the maximum likelihood state of speech frame, the effects of other suboptimal states are neglected and some important information is lost, which reduces the system recognition rate. Acoustic state likelihood modelling and supervised state modelling are used here to better utilize the acoustic state likelihood information. A state likelihood cluster feature and a supervised state feature are used to calculate the state likelihood of the acoustic feature Mel frequency cepstrum coefficient (MFCC). Tests show that these three features improve the speech recognition accuracy. The state likelihood cluster and supervised state feature reduce the relative error rate by 6.10% and 9.66% for isolated word recognition compared to GMM-HMM using only MFCC and by 2.53% and 11.05% for continuous speech recognition.

The torsional vibration of hybrid vehicle powertrain system not only relates to the safety of the vehicle, but also affects passenger comfort. Aiming at the demand for simulation modeling in torsional vibration research, a method is proposed to recognize the dynamic parameters and to correct engine torque fluctuation. First, dynamic models and their applicability of the main components in the powertrain system are briefly introduced. The focus is put on the constitution and characteristics of engine torques calculated by a single-cylinder engine model, and a theoretical formula is given for correction. Then, possible acquisition methods and credibility of the obtained dynamic parameters of the system are reviewed. Taking advantage of the characteristics of a hybrid power system, a method with implementation steps is proposed to identify these parameters by combining experiments and simulation. Finally, a series-parallel powertrain for city bus is taken as an example, where the applicability of the proposed method and the simulation model are verified by means of bench tests under the typical working conditions.

Fuel cell vehicles, an important class of new energy vehicles, use compressors to increase the inlet pressure into the fuel cell. The high efficiency and fast response of centrifugal compressors make them an ideal option for the compressors used in fuel cell vehicles. However, centrifugal air compressors have poor off-design performance. Flow field deviation analyses were used here to study the three-dimensional flow distributions inside a centrifugal air compressor for design and off-design conditions. The flow separation on the suction surface of the main blade for off-design conditions is shown to be the main factor leading to the poor performance. The pressure distribution change on the suction surface of the main blade between the off-design and design conditions is then used as the optimization objective to improve the centrifugal air compressor design. The analyses consider various back sweep angles, forward lean angles and vaneless diffuser lengths. The optimized air compressor efficiency is improved by 2% at the design working condition and 5% at the near-stall working condition. Thus, this optimization method using flow field deviation analyses is proven to be effective.