Search

Ultrasonic Lamb wave is an important measure for nondestructive testing and imaging of defects or damages within plate structure material. In this field, the damage index probability and delay-and-sum are two commonly used imaging methods. The first one can identify multiple defects with low localization accuracy, however, the second one has high localization accuracy only in the case of single defect. By combining the advantages of the two methods, this paper proposed a novel damage index weighted delay-and-sum (DIWDAS) imaging method based on time reversal Lamb wave to solve the problem of accurate localization of multiple defects in plat structure with sparse ultrasonic sensors. Experiments on samples of tested metal plates with one and two defects manifested the effectiveness of the proposed method.

In computer vision area, extracting effective information from video data to realize automatic human action recognition has become one of the most active research points. In order to deal with the high dimensional challenge from massive amount of video data and extract effective features by feature self-learning model based on deep learning, a method based on a sparse long-short term memory (LSTM)auto-encoder and improved 3D convolution neural network (3D CNN)is proposed in this paper. We first construct a sparse LSTM auto-encoder to extract the key frames. And then, to extract lower dimensional features for approximating the upper limit of these features classification ability, the 3D CNN model is improved by combing with the feature engineering method, that is recursive feature elimination (RFE)algorithm. Finally, we extract high-level features by the improved 3D CNN for human action recognition. The experimental results on typical datasets UCF101 demonstrate that our algorithm has significant effect on data dimensionality reduction and outperforms some competing hand-craft feature and single feature methods.

Depth completion is an important research field in self-driving which complements depths mapped from point cloud of Lidar. This paper shows a depth completion method combining traditional image processing and image optimization. In our method, a depth map is completed through a series of well-designed morphological dilation and filtering methods, and then is optimized referring to a RGB image and a confidence map. The method is simple, data independent and runs only relying on the CPU. It is evaluated on the challenging KITTI depth completion benchmark [20]. The result performs as good as IP-Basic and better than sparse CNN in depth accuracy. Furthermore, it optimizes the edges of objects in the depth maps, which has a greater help for image segmentation, obstacle perception or other tasks in self-driving.

Moving object detection from video data is one of the most active research points in computer vision area. Recently the $\ell$ 2,0 -norm has shown remarkable advantages in signal processing, and the total variation (TV) has been widely used in image processing due to its ability to preserve sharp edges, which are usually the most important for clear object extraction. In this paper, firstly, a unified regularization framework named $\ell$ 2,0 mixed norm and total variation regularized RPCA (MNTV-RPCA) is presented to extract the foreground, which is able to enhance the smoothness of boundary. Secondly, the alternating direction method of multipliers (ADMM) is employed to deal with the proposed constrained minimization problem. Experimental results show that our method have better performance than some state-of-the-art methods especially for the cases with dynamic backgrounds.

Human action recognition is one of the most active research areas of computer vision. With the rapid development of deep learning, using neural networks to realize action recognition becomes a popular thesis. This paper proposes a self-learned action recognition method based on neural networks. The proposed method trains dictionaries with sparse autoencoder (SAE) and extracts the key frames with sparse long short term memory (LSTM) autoencoder. Moreover, the testing samples in the dataset can be represented as a vector by the dictionaries for classification. This method can reduce data dimension effectively. Meanwhile, it is also applicable for action recognition. The experimental results on the DHA dataset prove the effectiveness and practicability of the proposed method.

This paper mainly studies a new road curvature estimation method for self-driving vehicles. A new contour-based clustering and the parallel-based voting mechanism are used to detect the lanes. And curve fitting with least square error method is used to fit the lane models. Compared with other methods, this method can estimate the road curvature and width in advance to help a vehicle pass the curve in a reasonable speed and a proper turning angle.

Human action recognition is a challenging and active research area in computer vision. In this paper, we propose a simple yet effective method, called the locality-constrained linear coding (LLC) based two-dimensional spatial-temporal templates, to learn a discriminative representation for human action recognition. Our proposed method calculates twodimensional spatial-temporal templates from each human action sequence as the global features to describe the human action information. To describe the local detailed features better, we construct a multi-layer patches descriptor by spatial pyramid matching (SPM) method. And we encode the patches descriptor by using LLC algorithm to obtain a coding with underlying properties of better construction and local smooth sparsity for human action recognition. To evaluate the proposed method, we evaluate and compare our algorithm with some state-of-the-art methods on both Weizmann and DHA datasets. Experimental results show that our method outperforms some state-of-the-art methods.

Both the plant parameter4 and delay time of the system may vary in the time-varying system, which makes system identification relatively complex. In this paper, a new system identification algorithm based on subspace method and cross-correlation function is proposed to identify system parameter and time-delay simultaneously. The time delay is identified with the calculation of the cross-correlation function between pre-processed input data that processed through initial plant parameters and original output data. Then, the time delay in the output data can be deleted with the time delay identified before and the parameters can be achieved by subspace identification algorithm. Therefore, the process parameters and time delay identification can be performed online by through above two alternative operations. The validity of the proposed identification algorithm is verified by the simulation study.

Fault isolation is an essential step in multivariate statistical process monitoring, which discovers the cause of the detected fault. The common adopted techniques for fault isolation, such as contribution plots, reconstruction analysis, branch and bound, etc., often suffer from the problems of the smearing effect, insufficient historical fault data, and heavy computational burden. In this paper, the nonnegative garrote (NNG) estimator is utilized for multivariate statistical fault isolation, which solves the problems of the existing methods. The proposed method is based on the finding that there is equivalence between fault isolation and variable selection in discriminant analysis. Different from the existing methods, the NNG-based fault isolation method does not suggest a single set of faulty variables. Instead, it provides a sequence of process variables according to their responsibilities for the detected fault. Such results are informative for diagnosing the root cause of the fault.

Pulsed thermography (PT) is a widely used non-destructive testing (NDT) technique for defect detection in carbon fiber reinforced polymers (CFRP) structures. Thermographic signal reconstruction (TSR) is often employed to process thermographic data due to its excellent performance in enhancing the contrast between defects and background. However, TSR does not reduce the noise of thermographic data effectively. To solve this problem, this thesis proposes an image denoising method based on wavelet transform. The experiment results show that dealt with wavelet transform the characteristics of the defective regions are more clearly compared with previous images. And the signal-to-noise ratio (SNR) values are improved obviously at the same time.

In industrial process, some important variables such as quality index, efficiency index and concentration of product components are difficult or even impossible to be measured directly due to the limitation of technology. This phenomenon leads to few labeled data and plenty of unlabeled data. Traditional identification method for controlled auto regressive (CAR) model usually cannot deal with unlabeled training data. As a result, these traditional identification methods may receive poor identification precision or even cannot work entirely. To solve the problems above, this paper proposes a new identification method based on deep learning (DL). Firstly, the CAR model is transformed into finite impulse response (FIR) model solve the problem of lack of autoregressive part; Secondly, autoencoder of deep learning make full use of unlabeled data to pretrain the model; Thirdly, small amount of label data is used for fine-tuning. As a semi-supervised learning method, deep learning can be able to extract more information from unlabeled data than traditional supervised learning method. The results show that the proposed method can acquire higher identification accuracy than BP neural network.

Based on the actual data analysis, it is found in this paper that the supplementary air flow rate in the CDQ operation didn't follow the variation of the discharge rate of incandescent coke well, which results in the concentration increase of combustible gas in the exhaust gas and the decrease of economic efficiency. The correlation analysis results show that the introduced derived variables are more useful than some plain variables for the purpose of prediction. Next, to handle the contradiction between the steam productivity and the coke burning loss, a new economic efficiency index is introduced by synthesizing the two competing aspects. A kind of soft sensor of economic efficiency is put forward by combining nonnegative garrote (NNG) variable selection algorithm with the autoregressive integrated moving average (ARIMA) model, which gives a good solution to the economic efficiency real-time prediction problem of CDQ system. Then, the implementation of model-based optimization is studied based on the actual operation data. The results show that there exists large room for economic efficiency promotion.

How to operate a BFG/coal co-firing boiler in high efficiency is challenging for a gas/solid multi-fuel combustion system. Taking operation data from a real boiler, this study proposes an operation optimization strategy of BFG/coal co-firing boiler based on deep learning. Firstly, the thermal efficiency model is constructed based on deep learning with all the actual sampling data, which outperform the other modeling methods. Then, the high efficiency data being selected as labeled data, a deep learning-based model of air flow is constructed. It needs to be emphasized that, to take full advantage of all available information and improve the model accuracy, all the labeled and unlabeled data are utilized in auto-coder learning of the air flow modeling. Finally, the constructed air flow model is used for the operation optimization and the thermal efficiency model is used to evaluate the level efficiency improvement. The high accuracy of the proposed modeling approaches makes the implementation of the operation optimization strategy readily practicable.

Heat management is one of the major challenges in modern electronic devices. The higher performance results in a production of greater amount of heat which needs to be efficiently dissipated so as to ensure the electronic devices operational during the period of lifetime. This paper discusses the application of micro- and nano-materials in thermal interface materials (TIM) used for heat management. Effects of type, size and geometry of different fillers were experimentally investigated. The results showed that it is recommended to utilize silver particles compared to copper ones to achieve higher heat dissipation. And the particles of smaller size may enhance the thermal conductivity of elaborated materials.

A theoretical method is proposed to evaluate the creep damage in a 91 steel by using ultrasonic nonlinear technique. A 2-dimensional finite element model is developed to simulate the creep damage of steel P91 and the ultrasonic wave propagation through the degraded material. The micro-structural changes in the degraded material cause the distortion in the ultrasonic wave, and this material's nonlinearity leads to the generation of the second-harmonic wave component. Both the experimental and theoretical results show that the material nonlinearity increases with the increase of thermal treatment temperature. The results show the feasibility of the model to evaluate the early stage of the degradation of materials by using ultrasonic nonlinearity.

The K-SVD algorithm is a powerful tool in finding an adaptive dictionary for a set of signals via using the sparse representation optimization and constrained singular value decomposition. In this paper, we first review the original K-SVD algorithm as well as some sparse representation algorithms including OMP, Lasso and recently proposed IITH. Secondly, we embed the Lasso and IITH sparse representation algorithms into the K-SVD process and establish two new different K-SVD algorithms. Finally, we have done extensive experiments to evaluate the performances of these derived K-SVD algorithms with different pursuit methods and these experiments show that the K-SVD with IITH has distinctive advantages in computational cost and signal recovery performance while the K-SVD with Lasso is not sensitive to initial conditions.

Nanomachine is a kind of nanoscale device, with the size ranging from one to a hundred nanometers. The nanomachines exchange information with each other by molecular communication, which is a novel bio-inspired communication technology. The synchronization among the nanomachines is important for a lot of applications in the development of instrumentation and measurement technologies at nanoscale. In this paper, a reference broadcast synchronization scheme for the nanomachines is proposed. In this scheme, a reference nanomachine sends a beacon to its child nanomachines. The child nanomachines exchange information to achieve the synchronization. The Gaussian distribution is used to model the molecular propagation delay between the nanomachines. The relative clock offset and the relative clock skew are estimated by joint maximum likelihood estimation. The simulation results demonstrate that the proposed synchronization scheme is effective.

Molecular communication is a very promising research field. It has a broad range of prospective potential applications in biomedical engineering, material manufacturing, etc. Getting the knowledge of the diffusive molecular channel is important for the design and analysis of the molecular communication system. This paper investigates the channel parameter estimation based on a typical inverse Gaussian distributed channel. The estimator of propagation distance, medium velocity and diffusion coefficient are derived by maximum likelihood estimation method. The simulation results validate the effectiveness of the proposed estimators.

This paper introduced the cloud theory briefly firstly, then presents a method based on the cloud theory to access the collective harmonic influence of the residential loads, which can identify the harmonic sources so as to determine the harmonic suppression scheme. For the residential harmonic current probability distribution has multiple peak values, the cloud transformation method has more advantages than the normal distribution method. The analysis results of practical cases prove that the cloud transform is capable of fitting the harmonic current distribution curves in high precision, and presenting the multi-modal characteristics of the distribution curves well.

The results of investigations of new compositions of electrically conductive adhesives (ECA) for printed electronics applications with and without nano-additives are presented. The studies covered the influence of nano additives on mechanical properties and reliability of interconnects made by elaborated, improved composition of the ECA based on primary conclusions. The obtained results showed that nano additives influence significantly and positively on mechanical properties of ECA joints, but their impact on the reliability properties of joints manufactured in low temperatures is much lower or even negative.

With the minimization development of electronic devices and products, nanotechnology and nanomaterials are widely applied in different fields of electronic packaging. Carbon nanotube (CNT) is an ideal material due to its excellent electrical and thermal conductivities. In the present paper, the application of CNT bundles as chip bumps was experimentally investigated. The electrical resistances of the CNT interconnects were measured, and the thermal and humidity test were conducted. In addition, the CNT forests on fine pitch copper lines under various environmental test conditions were observed to evaluate the stability.

The BOA framework in which BOM is the basis has attracted much attention in developing ERP system of SMEs for solving the problem of information combination. In this paper, in order to provide VSP in BOA framework with large grain reusable module, to improve developing efficiency and reduce development and maintenance costs of ERP systems, a BOM based reconfigurable developing method is presented. According to the idea of BOA and convergent evolution, a design principle to develop BOM by using the approach of meet-in-the-middle formatting is proposed. A rapid VSP developing process by managing BOM and configuring BOM instances has also put forward. At the end of the paper, the feasibility of the reusable development method is illustrated with an application example.

The paper presented the results of investigation of two solder pastes SnBi and SAC with different nano-particles, which could be applied for special application. It was presented that materials and samples preparation for investigation. The results of SEM analyzer of solder joints executed using these pastes and their shear strength measurements enable to make the comparison of nano-additives influence on properties of the bi-modal solders pastes. It was observed that addition of nano-particles into both solder pastes changed the microstructure of solder joints, and formed a large number of nano-sized grains with uniform distribution. For the SAC solder joints with nano-Ag particles it was also observed that specific Ag3Sn crystallites. We supposed that the Ag3Sn crystals, which created "internal net" and nano-sized grains with uniform distribution must influence the mechanical properties of the solder joints. The results of shear strength measurements of solder joints confirmed the significantly higher shear strength of bimodal solder joints in comparison to reference solder joints without nano-additives.

An improved face detection and recognition method based on information of skin color and depth obtained by binocular vision system was proposed in this paper. With this method, the face area was detected firstly by using Adaboost algorithm. Afterwards, the real face was distinguished from fake one by using the skin color information and the depth data. Then, by using PCA algorithm, a specific face can be recognized by comparing the principal components of the current face to those of the known individuals in a facial database built in advance. This method was applied to a service robot equipped with a binocular camera system for real-time face detection and recognition experiment, and satisfactory results were obtained.

An improved Hilbert-Huang transform(HHT) combined with wavelet packet transform(WPT) is proposed for recognizing continuous electroencephalogram(EEG) in brain computer interfaces(BCIs). The HHT consists of empirical mode decomposition(EMD) and Hilbert-Huang spectrum(HHS). Firstly, the WPT decomposes the signal into a set of narrow band signals, then a series of Intrinsic Mode Functions(IMFs) can be obtained after application of the EMD. Whereafter, two kinds of screening processes are conducted on the first two IMFs of each narrow band signal to remove unrelated IMFs. Hilbert Transform(HT) is then employed to calculate the HHS, from which energy changes in mu-rhythm and beta-rhythm can be recognized clearly. Datasets I of BCI competition IV 2008 are analyzed. The results show that the proposed method has better discriminability than the traditional HHT among different states. The proposed algorithm has the potentiality to trace mu-rhythm and beta-rhythm changes, which paves a way for a more enhanced BCI performance.

Modular inversion and Scalar multiplication are the kernel for computations in the GF(p) used by many public-key cryptosystem in data security systems. Various algorithms have been used for both software and hardware implementation of GF (p) inversion, but its application is limited because of the constraint of its speed. In this paper, a fast GF(p) inversion algorithm is proposed which can be used in both software and hardware realizations and is particularly suitable for VLSI implementations. Also an efficient algorithm for scalar multiplication in ECC is discussed and the corresponding comparative results are given. The results of numerical tests and performance comparisons manifests that the proposed algorithms can remarkably improve the computational efficiency. Hence, it has practical significance for the implementation of ECC and is expected to be applied to data security.

Anisotropic conductive films (ACFs) are composed of an adhesive polymer matrix and fine conductive fillers using metallic particle or metal-coated polymer balls. Flip chip interconnection using ACF is being widely used in the electronic packaging. In this paper, two-dimensional nonlinear finite element analysis model was built to simulate effects of bumps' metallization on the stress and strain distributions of ACF interconnections for flip-chip-on-flex (COF) applications. The simulation results show that the bump metallization has big influence on the stress and strain. Under the thermal mechanical simulation, the stress is mainly relied on the Young's modulus of the bump while the strain is mainly depended on the CTE of the bump. In some kinds of bump/pad metallization, such as gold, nickel and copper, the copper has smaller stress, only bigger than gold, and minimum strain. If both chip bump and substrate bump fabricated by same copper, the galvanic corrosion, which occurs between two different metals and the main cause attributed to the shift of the contact resistance of ACF, does not exist. Regarding the cost and reliability synchronously, the copper may be the better substitute of gold as bump material in flip chip using ACF

Ultrasonic echoes, which contain useful information related to material micro-damages in nondestructive evaluation, usually exhibit involved time-varying frequency properties. In this paper, continuous wavelet transform (CWT) based time-frequency analysis for echo signal were discussed. Ultrasonic testing of pulse-echo technique with high frequency contact transducers was carried out to thermally degraded Cr-Mo steel samples processed by different high temperature heat treatment conditions. Obtained echo signals were analyzed by CWT based on Gabor wavelet. Two major frequency dependent ultrasonic parameters, attenuation and velocity, were obtained from the local peak values at each frequency from the magnitude of CWT. Further, numerical correlations of the attenuation coefficients and group velocity with the sample's internal grain size were obtained. This method has a potential for quantitative evaluation to damage advance before crack initiation

The first practical light emitting diode (LED) was created almost four decades ago. With the explosion of the internet, mobile phones, portable laptop computers, and a myriad of other electronic items, we are now in the middle of an electronic revolution. LED is playing an important role in this revolution because of the numerous fields in which LEDs are used such as lights, displays, projectors, even vision aid, etc. Advances in the development of LEDs have now made LED semiconductor components available across the entire color spectrum. In much the same way as fluorescent lighting forty years ago, LED revolutionizes the entire field of lighting technology. Achieving maximum light yield at minimum production cost is a decisive factor for new low-cost applications. Thus flip chip (FC) technology is attracting rapidly increasing interests from research organizations to relevant companies. Due to flip chip design all electrical contacts are formed in one assembly step and wire bonding isn't necessary. The second advantage is the good heat transfer that allows high current and therefore high power of the LED. And another advantage of flip chip is that it has a much bigger lighting surface than traditional packaged LEDs, this will make the LEDs hold a high brightness and lighting efficiency.

In this paper, a novel double-T-shaped microstrip resonator is presented, and its electromagnetic field distribution, resonant frequency and quality factor are computed and analyzed. A novel double-T-shaped microstrip bandstop filter is designed and its filter performance is simulated and analyzed as well. The results show that the filter has outstanding virtues such as wide bandwidth, high attenuation and no parasitical passband in stopband. The filter can be applied in RF circuit for its simple structure, easy manufacture, low cost