Publications

This paper presents element level structural damage quantification using an ensemble-based machine learning technique, namely, random forest technique, with acceleration responses from structures. The ensemble-based approach provides a better prediction than an individual model. Random forest is a machine learning algorithm which has several decision trees to perform a task. The proposed approach develops a random forest as a regressor to predict multiple output variables, which is the vector of elemental level damage quantification results of the structure. Damage severity is identified as the reduction in the elemental stiffness parameters. The acceleration responses for single-element and multiple element damage cases are generated and further processed to feed as input to the random forest. The acceleration responses from the sensor nodes are concatenated, and principal component analysis (PCA) is applied to reduce the uncorrelated input dimension. The proposed approach can provide good damage identification results efficiently with much less computational demand and time, compared to the neural network training methods and deep learning models. Moreover, fewer sensors are needed to measure acceleration responses to localise damage and quantify the damage level. To demonstrate the proposed method, a simply supported beam is used as an example in numerical studies. Different levels of noise in the acceleration responses and uncertainty in the finite element modelling are considered. Experimental studies on a steel frame structure with 70 elements are also conducted to investigate the performance of the proposed approach for structural damage quantification. Good identification results are obtained with an efficient training process.

A back analysis model is established for analyzing the thermal characteristics parameters of concrete, and an optimized back analysis process is programmed by using FEM simulative calculation and fixable tolerance method. Numerical examples shows the reliability of the calculation method, and it indicates a relatively rapid convergence. A project example is given in the paper for illustration. The obtaining of thermal characteristics parameter of concrete by optimized back calculation on the basis of site temperature observation values will become one of the effective way to partially or totally replace the parameters obtaining only from laboratory experiments.

In this paper, the different sizes of enlarged weld access hole (EWAH) in steel beam-column connection are presented. Based on the commercial nonlinear finite element code ABAQUS, firstly, the sequential welding process is simulated, and the non-linear pseudo-static is performed. Through the comparison of results between the FEA and finished pseudo-static experiment when EWAH size is adopted by x =115mm and y =35mm, it is demonstrated that this analysis method is rational and effective, moreover, the stress concentration effects and cracks are discovered. And then the FEA method still obtains, the EWAH size y =35mm is changeless, the nine structural models are made up for only considering the value range from x =70mm to x =110mm. Furthermore, the impacts of the EWAH size on mechanical behaviour are discussed, and the reasonable geometric range is proposed.

Inspired by the progress of the End-to-End approach [1], this paper systematically studies the effects of Number of Filters of convolutional layers on the model prediction accuracy of CNN+RNN (Convolutional Neural Networks adding to Recurrent Neural Networks) for ASR Models (Automatic Speech Recognition). Experimental results show that only when the CNN Number of Filters exceeds a certain threshold value is adding CNN to RNN able to improve the performance of the CNN+RNN speech recognition model, otherwise some parameter ranges of CNN can render it useless to add the CNN to the RNN model. Our results show a strong dependency of word accuracy on the Number of Filters of convolutional layers. Based on the experimental results, the paper suggests a possible hypothesis of Sound-2-Vector Embedding (Convolutional Embedding) to explain the above observations. Based on this Embedding hypothesis and the optimization of parameters, the paper develops an End-to-End speech recognition system which has a high word accuracy but also has a light model-weight. The developed LVCSR (Large Vocabulary Continuous Speech Recognition) model has achieved quite a high word accuracy of 90.2% only by its Acoustic Model alone, without any assistance from intermediate phonetic representation and any Language Model. Its acoustic model contains only 4.4 million weight parameters, compared to the 35~68 million acoustic-model weight parameters in DeepSpeech2 [2] (one of the top state-of-the-art LVCSR models) which can achieve a word accuracy of 91.5%. The light-weighted model is good for improving the transcribing computing efficiency and also useful for mobile devices, Driverless Vehicles, etc. Our model weight is reduced to ~10% the size of DeepSpeech2, but our model accuracy remains close to that of DeepSpeech2. If combined with a Language Model, our LVCSR system is able to achieve 91.5% word accuracy.

Impact brings great threat to the composite structures that are extensively used in an aircraft. Therefore, it is necessary to develop an accurate and reliable impact monitoring method. In this paper, fiber Bragg grating (FBG) sensors are embedded in unidirectional carbon fiber reinforced plastics (CFRPs) during the manufacturing process to monitor the strain that is related to the elastic modulus and the state of resin. After that, an advanced impact identification model is proposed. Support vector regression (SVR) and a back propagation (BP) neural network are combined appropriately in this stacking-based ensemble learning model. Then, the model is trained and tested through hundreds of impacts, and the corresponding strain responses are recorded by the embedded FBG sensors. Finally, the performances of different models are compared, and the influence of the time of arrival (ToA) on the neural network is also explored. The results show that compared with a single neural network, ensemble learning has a better capability in impact identification.

Accurate State-of-Charge (SOC) estimation and reasonable charge equalization for power batteries is important for prolonging their lifetime. In the application of electric agricultural machinery, the SOC estimation of Ni-MH power batteries was implemented by the Square-Root Central Difference Kalman Filter (CDKF) approach in this paper. To provide charge equalization for all the battery strings in the test module, a relay matrix module was used for the power converter and the equalization method was based on the estimated SOC of the battery strings. The test results confirm that the SOC obtained from the Square-Root CDKF method can converge to the true biases much quicker in comparison with the EKF method. The experimental data illustrate that the battery strings are balanced well at the end of the charge equalization mode.

For the study of characteristics of the lightning around the capsized "Oriental Star" incident, this paper analyzes the lightning space distribution in the surrounding, preliminary judging from characteristics of the lightning figure out the form of disastrous weather patterns.It is concluded that: the wreck accident happened in lightning highest frequency period, also most lightning frequency region, as well as the biggest negative rather than positive lightning current time bucket and region.The latest lightning is less than 300 m from the accident, leads to a higher possibility of damage to the ship communications equipment by lightning electromagnetic induction, which needs to further strengthen the safety precautions against ships during a thunderstorm.

<title>Abstract</title> Neurogenic lower urinary tract dysfunction (NLUTD) is a frequent consequence of spinal cord injury (SCI), leading to symptoms that significantly impact quality of life. However, existing treatment strategies for managing NLUTD exhibit limitations and drawbacks. The demand for a novel, effective approach to restore bladder function and re-establish urination control. In this study, we introduce a new electrical neuromodulation strategy involving electrical stimulation of the major pelvic ganglion (MPG) to initiate bladder contraction, in conjunction with innovative programmable (IPG) electrical stimulation on the pudendal nerve (PN) to induce external urethral sphincter (EUS) relaxation in freely moving or anesthetized SCI mice. Furthermore, we conducted the void spot assay and cystometry coupled with EUS electromyography (EMG) recordings to evaluate voiding function and monitor bladder pressure and EUS muscle activity. Our findings demonstrate that our novel electrical neuromodulation approach effectively triggers coordinated bladder detrusor contraction and EUS relaxation, effectively counteracting SCI-induced NLUTD. Additionally, this electrical neuromodulation method enhances voiding efficiency, closely resembling natural reflexive urination in SCI mice. Thus, our study offers a promising electrical neurostimulation approach aimed at restoring physiological coordination and potentially offering personalized treatment for improving voiding efficiency in individuals with SCI-associated NLUTD.

In these days,multi-touch technology becomes an important HCI device,it is used in various environments.This paper proposes an infrared multi-touch recognition algorithm model,which uses two different bi-axial systems to recognize multi-touch of users;it solves the multi-touch recognition problems in traditional infrared touch screen,the comparing result shows that this algorithm improves the recognition rate of infrared multi-touch screen.

The method of long range network RTK is the main means to improve the accuracy of real-time kinematic positioning of GPS/BDS. The fast fixing of carrier phase integer ambiguity with GPS/BDS between long range reference stations was significant. The algorithm of multi-frequency carrier phase integer ambiguity resolution with GPS/BDS between long range network RTK reference stations was studied in this paper. The GPS wide-lane ambiguities were fixed by the double frequency observation data of reference station network. And the extra-wide-lane ambiguities of B2 and B3 frequency were fixed by using B2 and B3 frequency observation data at the same time. Then the solution model including GPS double difference carrier phase integer ambiguities and atmospheric errors can be established, the carrier phase integer ambiguity can be fixed by this solution model and the constraint of wide-lane ambiguities. Then the spatial correlation model of atmospheric delay error was established. And the computation model of BDS double difference carrier phase integer ambiguity including atmospheric delay error was established by the extra-wide-lane ambiguities of B2 and B3 frequency. The BDS carrier phase integer ambiguity resolution was constrained by the spatial correlation model of atmospheric errors. Therefore the influence of ambiguities fixing by using ionosphere free observation can be eliminated. The test of the algorithm was carried out by the GPS/BDS data with long range reference station network. The results of experiment indicate that the rapidly fixing of carrier phase integer ambiguity with GPS/BDS between long range reference stations was achieved by this algorithm.

Abstract Reliability analysis of bridges is essential for the design of civil engineering structures. The classical methods, such as Monte Carlo Simulation (MCS) and subset simulation techniques, may provide accurate results. However, since the finite element model of the large‐scale civil engineering structures usually consists of a large number of degrees of freedom, structural reliability analysis of such structures is time‐consuming and computational intensive, which may restrict the use of these methods. This paper proposes a novel method for the reliability analysis of bridge under different types of loads by combining Polynomial Chaos (PC) and subset simulation techniques. The surrogate model of the limit state function is approximated by using the PC expansion, in which the PC coefficients are obtained from the least‐squares method. The subset simulation with a PC‐based surrogate model is then used to estimate the rare failure probability. Reliability analyses of bridge structures under different loading types, that is, static load, dynamic moving load, and seismic load are performed. Studies by using the MCS method and the classical subset simulation are also conducted, and the results from the proposed approach are compared with those from MCS and subset simulation to demonstrate the accuracy and efficiency of the proposed method.