Search

The Fan-out Panel Level Packaging model after the molding process is established. We simplify the complex curing reaction of epoxy molding compound with equivalent thermal expansion coefficient method. We successfully use finite element method to build different geometric Fan-out Panel Level Packaging model as training dataset for machine learning. In the end, we used the modified Convolution Neural Network to estimate the amount of fan-out panel-level package warpage under different geometries. To avoid local minimum, Particle Swarm Optimization, PSO, is introduced for neural network parameter.

The United States creates or acquires increasingly more complex intelligence, surveillance, and reconnaissance (ISR) systems to maintain a strong, leading presence within the world. As a result, ISR systems have become more costly and difficult to manage. The research team focused on continuing previous year efforts of another team to utilize commercial off-the-shelf (COTS) technologies in the development of more flexible and cost-effective ISR systems. The primary goal was to design and implement an autonomous quadcopter that integrated an Android smartphone, an Arduino microcontroller, and several ultrasonic sensors to independently explore and map an unknown area. The project was broken down into three main tasks: construction of the quadcopter and integration of ultrasonic sensors, Android phone, and Arduino microcontroller, development of an Android application that generates navigation commands and avoids collisions, and development of an Android application that uses sensor data for Simultaneous Localization and Mapping (SLAM). This project, a proof-of-concept of a quadcopter system for autonomous navigation and mapping of an unknown environment, demonstrates the feasibility of developing inexpensive ISR systems with commercially available products. The team also found that the Arduino-Android interface was quite complex and caused issues with basic flight stability. The team also found that ultrasonic sensors were capable of partial SLAM by producing rudimentary maps under controlled conditions and simulated stable flight. However, the inexpensive sensors are unlikely to yield the detailed maps necessary for autonomous flight or actionable navigation information.

We have experimentally investigated dual-layer fishnet structures in the terahertz regime. The dispersion behavior of the different modes in this structure is inferred from their response to changes in THz wave incident angle and the thickness of the dielectric layer separating the two metallic films. The theory of magnetic plasmon polariton (MPP) is applied to explain the coupled dispersive modes.

As one of the most important characteristics of structural steels, toughness is assessed by the Charpy V-notch impact test. The absorbed impact energy and the transition temperature defined at a given Charpy energy level are regarded as the common criteria for toughness assessment. This paper aims at establishing generic toughness prediction models which link materials compositions and processing conditions with Charpy impact properties. Hybrid knowledge-based neural-fuzzy modeling techniques which incorporate linguistic knowledge into data-driven neural-fuzzy models have been used to develop the Charpy properties prediction models for thermomechanically controlled rolled (TMCR) steels. Two basic ways of knowledge incorporation are introduced to improve the performance of the obtained fuzzy models. Simulation experiments show that both numeric data and linguistic information can be combined in a unified framework and that both Charpy impact energy and the impact transition temperature (ITT) can be predicted by the same model

Studies of multi-millijoule pulse generation revealed ultimate peak-power generation limits set by bulk damage and self-focusing in large-core fibers. Ultrashort pulse or narrow-linewidth amplification is more stringently limited by Raman and Brillouin scattering, and self-phase-modulation.

An adaptive fuzzy control algorithm and corresponding rule generating approach are presented. According to this paradigm, the fuzzy control rules can be generated automatically from the input data, and then the appropriate control actions can be deduced efficiently by a simplified fuzzy inference engine. Compared to traditional self-organising fuzzy controllers, this method has less computational burden and faster reasoning speed.

An on-line information processing method, called characteristics recognition (CR), which is based on symbolic manipulation, phase-plane analysis and fuzzy classification is proposed for complex process control systems in which there are no reasonable mathematical models available. Emulating human decision processes which apply on-line characteristic information to solve complex control problems, this paper presents an alternative to extract and recognise the characteristic patterns which can capture system behaviour features and monitor control performance without prior process model. The key ideas, performance feature analysis in normalised phase-plane, characteristic pattern classification and recognition method are discussed in this paper.

The authors have developed a handwritten word recognition scheme based on a single contextual, discrete symbol probability hidden Markov model (HMM) incorporated with an adaptive length Viterbi algorithm. This work attempts to extend the earlier HMM scheme for naturally segmented word recognition to cursive and nonsegmented word recognition. The algorithm presegments the script into characters and/or fractions of characters, dynamically selects the correct segmentation points, determines the word length, and recognizes the word according to the maximum path probability. The HMM is on top of, but independent of, script segmentation and character recognition techniques, and therefore leaves room for further improvement. The experiments have shown promising results and directions for further improvement. >

A bidirectional serial search pseudonoise (PN) acquisition scheme in a direct sequence spread spectrum system is investigated. By starting with a common initial phase point, both the forward sliding correlation and the backward sliding correlation are performed and examined simultaneously. Whichever phase point in either direction hits the correct phase point first is selected as the desired reference phase point. The author describes the analysis model of bidirectional serial search and the procedure for calculating the overall mean PN acquisition time. A formula for the overall mean PN acquisition time is derived. The computational result and the performance comparison between the bidirectional serial search and the unidirectional serial search are given. >