Search

The general flow characteristics of the molten steel and the behavior of molten steel and liquid slag interface (level fluctuation) in the mold with a new type of electromagnetic brake, freestanding adjustable combination electromagnetic brake (FAC-EMBr), was investigated by numerical simulation. The influences of the magnetic induction intensity, the submerged entry nozzle (SEN) immersion depth, the SEN port angle, and casting speed on the effect of FAC-EMBr are investigated. The results show that electromagnetic field formed by FAC-EMBr can cover the jet flow impact region, the upward backflow region, and the meniscus region simultaneously. With the increase of magnetic flux density, the jet flow impact intensity and the molten steel velocity in the upward and downward backflow regions are effectively suppressed, the meniscus wave height also decreases gradually, and the level fluctuation tends to be stable. For the influence of SEN port angle and SEN depth on the effect of FAC-EMBr mold, the results show that the molten steel velocity in the three main influence regions are all effectively depressed with increase of the value of SEN port angle and SEN depth. Even though under the condition of high casting speed (VC = 2.2 m/min), the application of FAC-EMBr can achieve effective braking effect.

The brake effect of the freestanding adjustable combination electromagnetic brake (FAC-EMBr) and EMBr ruler on the behavior of molten steel flow and the level fluctuation were investigated with the numerical method. The effects of the horizontal magnetic pole position (EMBr ruler), magnetic induction intensity, and casting speed on two types of electromagnetic brakes were studied. The numerical simulation results show that the magnetic field caused by the EMBr ruler is mainly distributed under the submerged entry nozzle (SEN), and it is very weak nearby the meniscus area. After the FAC-EMBr is applied, the magnetic field is mainly distributed in the area below the submerged entry nozzle, the upper roll region, and in the meniscus region. The application of the electromagnetic brake can effectively suppress the impact of the jet and decrease the molten steel velocity in the meniscus area. The brake effect of the EMBr ruler on the behavior of the molten steel flow and the level fluctuation is significantly influenced by the horizontal magnetic pole position. The increasing of the magnetic flux density can significantly increase the velocity of molten steel in the upper roll region and lead to an intense fluctuation in the steel/slag interface, as the horizontal magnetic field cannot cover the three key regions. The brake effect of the FAC-EMBr is less influenced by the variation of the process parameters due to the addition of vertical magnetic poles. Additionally, the &ldquo, secondary braking effect&rdquo, of the vertical magnetic poles can help to lower the increase of velocity in the upper roll region caused by the excessive magnetic induction intensity and the high casting speed. Therefore, even under the high casting speed conditions, the application of a new type of FAC-EMBr is also an efficient way to suppress the molten steel flow and level fluctuation at the meniscus area and decrease the possibility of slag entrapment.

To estimate time complexity for a given algorithm is important for algorithm designers. Usually, time complexity means the "analytical" time complexity which needs to be proved by strict math derivation. We propose to estimate the "numerical" time complexity (NTC), which measures the minimum number of operations an algorithm has to spend, as well as capture the intrinsic laws of time complexity. The unique challenges include: (1) How to make a machine learning model has the same ability as a real-world CPU (2) How to measure the minimum number of required arithmetic operations for a given problem. To tackle these challenges, we first propose a memory-based recurrent calculation network to mimic the functions of CPU and then we propose a self-adaptive selection gate for deciding when the mimic calculation process should stop. In addition, we use a symbolic learning method to find the time complexity formula. We train and test our model on four basic algorithms: long integer addition, 1-dim max-pooling, outer product, and sorting. Experiment results demonstrate that our model can precisely predict the numerical time complexity as well as the time complexity formula for each algorithm. We also conduct many visualizations to prove the effectiveness and correctness of our model.

Bar-like sand mould tilted angle reduced free surface fluctuations. The primary and secondary reflected wave heights decreased with the scaling of Re0.8 and Re0.55, respectively. These heights were reduced by 54% and 51% when θ was increased from 0° to 5°. A larger tilted angle was found to generate an air cavity in the cast, the size of which was observed to vary with Reynolds number (Re). An optimised range of tilted angle (2°–3°) was proposed as a compromise to reduce the free surface fluctuations and maintain a perfect casting shape at the end of the filling stage.

The steel/slag interface behavior under a new type of electromagnetic brake (EMBr), vertical electromagnetic brake (V-EMBr), was investigated. The influence of the magnetic induction intensity, the submerged entry nozzle (SEN) immersion depth, and the port angle of the SEN are investigated numerically. The effect of magnetic induction intensity on the meniscus fluctuation of molten alloy is further studied by the experiments. The results show that the meniscus fluctuation is depressed as the magnetic induction intensity is increased, especially for the region in the vicinity of the narrow face of the slab mold. This result is validated by the following experiments. For the influence of the SEN immersion depth and the port angle, the results show that the meniscus fluctuation is suppressed as the values of the immersion depth and the port angle are increased (absolute values for the port angle). However, the influence of the immersion depth and the port angle are not as sensitive as those in the other type of EMBr, e.g., EMBr Ruler. The industrial application of V-EMBr could benefit from this result.

We experimentally investigate the influence of the electromagnetic stirring (EMS) on the microstructure and the mechanical properties of Incoloy825 superalloy. Three trails of experiments with different values of the EMS alternating current (I = 0, 150 and 300 A, respectively), however with identical EMS frequency (f = 5 Hz), were conducted. We figure out that the equiaxed grain region is increased and the average grain size is decreased whilst EMS is applied. The application of EMS can also decrease the total number of TiN, especially for the inclusion in large scales (>2 μm). The values of the yield strength, the ultimate tensile strength and the elongation of Incoloy825 superalloy specimens are also increased whilst EMS is applied.

This research aimed at reveal the reasons for the extra Co–Cr–W alloy wastage in the risers in sand casting. The alloy filling behaviour in both the original and modified moulds was investigated numerically. The alloy–air interface was captured by using volume of fraction method. For the original mould, an unfilled volume in the vicinity of the runner bar top was apparent and it was refilled by a back flow, originated from the risers in the late stage of filling. The back flow behaviour required a higher level of the liquid alloy in the risers, which resulted in excessive wastage, and it was essential to form the required shape of the cast. For the modified mould, the unfilled volume was eliminated and the cast part shaping time was reduced to around 10 s from 90 s. The alloy wastage in the risers was reduced by 11%.

Deep-learning (DL) is receiving huge attention as enabling techniques for emerging mobile and IoT applications. It is a common practice to conduct DNN model-based inference using cloud services due to their high computation and memory cost. However, such a cloud-offloaded inference raises serious privacy concerns. Malicious external attackers or untrustworthy internal administrators of clouds may leak highly sensitive and private data such as image, voice and textual data. In this paper, we propose Occlumency, a novel cloud-driven solution designed to protect user privacy without compromising the benefit of using powerful cloud resources. Occlumency leverages secure SGX enclave to preserve the confidentiality and the integrity of user data throughout the entire DL inference process. DL inference in SGX enclave, however, impose a severe performance degradation due to limited physical memory space and inefficient page swapping. We designed a suite of novel techniques to accelerate DL inference inside the enclave with a limited memory size and implemented Occlumency based on Caffe. Our experiment with various DNN models shows that Occlumency improves inference speed by 3.6x compared to the baseline DL inference in SGX and achieves a secure DL inference within 72% of latency overhead compared to inference in the native environment.

This paper presents a generative adversarial networks (DUGan) for images registration, which is unsupervised without ground-truth deformations. In this registration framework, the generative network is in charge of generating the deformation field, which is the input to the transformation module, whose output is the wrapped images. The wrapped and fixed images pair will be judged the similarity by the discriminator network. The first innovation is that the U-Net model is introduced in the two networks making use of its hierarchical advantages. And the second one is that multiresolution loss computation is adopted to keep the deformation smoothness in the generative network, and guarantee the efficient similarity in the discriminator network. Experiments on brain datasets indicate that our method yields comparable accuracy to state-of-the-art registration methods.

In this paper we present a novel and general method to accelerate convolutional neural network (CNN) inference by taking advantage of feature map sparsity. We experimentally demonstrate that a highly quantized version of the original network is sufficient in predicting the output sparsity accurately, and verify that leveraging such sparsity in inference incurs negligible accuracy drop compared with the original network. To accelerate inference, for each convolution layer our approach first obtains a binary sparsity mask of the output feature maps by running inference on a quantized version of the original network layer, and then conducts a full-precision sparse convolution to find out the precise values of the non-zero outputs. Compared with existing work, our approach avoids the overhead of training additional auxiliary networks, while is still applicable to general CNN networks without being limited to certain application domains.

Neural networks based on Long Short-Term Memory (LSTM) are widely deployed in latency-sensitive language and speech applications. To speed up LSTM inference, previous research proposes weight pruning techniques to reduce computational cost. Unfortunately, irregular computation and memory accesses in unrestricted sparse LSTM limit the realizable parallelism, especially when implemented on FPGA. To address this issue, some researchers propose block-based sparsity patterns to increase the regularity of sparse weight matrices, but these approaches suffer from deteriorated prediction accuracy. This work presents Bank-Balanced Sparsity (BBS), a novel sparsity pattern that can maintain model accuracy at a high sparsity level while still enable an efficient FPGA implementation. BBS partitions each weight matrix row into banks for parallel computing, while adopts fine-grained pruning inside each bank to maintain model accuracy. We develop a 3-step software-hardware co-optimization approach to apply BBS in real FPGA hardware. First, we propose a bank-balanced pruning method to induce the BBS pattern on weight matrices. Then we introduce a decoding-free sparse matrix format, Compressed Sparse Banks (CSB), that transparently exposes inter-bank parallelism in BBS to hardware. Finally, we design an FPGA accelerator that takes advantage of BBS to eliminate irregular computation and memory accesses. Implemented on Intel Arria-10 FPGA, the BBS accelerator can achieve 750.9 GOPs on sparse LSTM networks with a batch size of 1. Compared to state-of-the-art FPGA accelerators for LSTM with different compression techniques, the BBS accelerator achieves 2.3 ~ 3.7x improvement on energy efficiency and 7.0 ~ 34.4x reduction on latency with negligible loss of model accuracy.

Electromagnetic brake (EMBr) technique is adopted to reduce the turbulence of molten alloy in the slab mould in the continuous casting process, especially under high casting speed. We introduce a state-of-the-art EMBr technique by reviewing the published literature. The main objective of this paper is to give a clear view of the EMBr technique in terms of the magnet arrangement, along with their “Braking” effect to help decision-making. The EMBr system can be divided into three types, in terms of the magnet arrangement: the Local type, the Ruler type and the Multi-mode type, respectively. Both advantages and disadvantages of each type have been discussed. Further challenges are also raised.

Existing task-oriented dialogue systems seldom emphasize multi-intent scenarios, which makes them hard to track complex intent switch in a multi-turn dialogue, and even harder to make proactive reactions for the user’s next potential intent. In this paper, we formalize the multi-intent tracking task and introduce a complete set of intent switch modes. Then we propose ISwitch, a system that can handle complex multi-intent dialogue interactions. In this system, we design a gated controller to recognize the current intent, and a proactive mechanism to predict the next potential intent. Based on these, we use pre-defined patterns to generate proper responses. Experiments show that our model can achieve high intent recognition accuracy, and simplify the dialogue process. We also construct and release a new dataset for complex multi-turn multi-intent-switch dialogue.

We investigate the spectral power distributions of 54 phones (33 measured experimentally in default text mode and 21 downloaded from the web) and estimate the mean ± std. dev. of the luminous efficiency of radiation, melanopic efficiency of radiation, and melanopic/photopic ratio as 287 ± 13 lm/W, 303 ± 26 blm/W, and 1.06 ± 0.13, respectively. We establish the chromaticity–performance characteristics relation to precisely assess the action efficiency of radiation performance using either RGB gray values or CIE xy values. Our real-time assessment of smartphone displays can aid in reducing energy consumption and improving user health.

The lack of labeled data is one of the main challenges when building a task-oriented dialogue system. Existing dialogue datasets usually rely on human labeling, which is expensive, limited in size, and in low coverage. In this paper, we instead propose our framework auto-dialabel to automatically cluster the dialogue intents and slots. In this framework, we collect a set of context features, leverage an autoencoder for feature assembly, and adapt a dynamic hierarchical clustering method for intent and slot labeling. Experimental results show that our framework can promote human labeling cost to a great extent, achieve good intent clustering accuracy (84.1%), and provide reasonable and instructive slot labeling results.

In this work, we synthesized a sun-like light source with tri-chromatic broad spectra, composed of light source units with single-peak 435-nm carbon quantum dots (QDs) and 695-nm perovskite QDs, and dual-peak 510- and 590-nm Ag-doped InP QDs, referencing CIE standard illuminants D50 and D65. Based on a spectral angle measure approach, we adopt the spectral similarity index (SSI) to measure the similarity between the synthesized light source and reference light source. We quantify the colour rendering and SSI of the tri-chromatic light source units and obtain the colour rendering and SSI as a function of chromaticity coordinate. The synthesized light source reaches an ideal performance with a maximum spectral similarity of 0.98 and colour rendering of about 93. This work is the first to synthesize a full-spectrum sun-like source with three QD light units, including a dual-peak QD unit, and proves QD LEDs as promising light sources for synthesizing full-spectrum light sources. In the future, improving the long wavelength part (>700 nm) of the broad-spectrum QD LED unit may significantly improve the colour rendering property.