基于YOLOV4的工件表面质量在线检测方法研究

目的 提升自动化产线上工件表面微小缺陷的检测精度和检测速度。方法 首先,在预处理阶段提出采用CutMix的数据增强方法,增加训练样本的多样性,提高模型的鲁棒性和泛化能力,避免训练模型产生过拟合;使用K–means++聚类算法生成边界候选框,以适应不同尺寸的缺陷,并较早地筛选出更精细的特征。其次,借助CSP Darknet53网络及SPP模块提取输入原始图像的特征,通过训练获得针对工件表面质量的在线检测模型,提升YOLOV4缺陷位置检测及识别的精度。结果 实验结果表明,文中所提出的基于YOLOV4的工件表面质量在线监测方法的预测精度达到97.5%,检测速度达到32.8帧/s,均优于同类的深度学习算法。以贵州某航空工业产品的自动化产线作为实验平台验证了所提方法的可行性和有效性。结论 该方法具备结构简单清晰、自适应性强等优点,检测精度和速度均满足工业场景需求,可以将其用于产品表面质量的在线检测。     

Frenet坐标框架下的自适应避障算法

目的 解决包装车间无人运输车辆在沿全局路径行驶过程中,难以同时保证跟随轨迹平滑及规避障碍物的问题。方法 构建Frenet坐标框架,描述车辆位姿与全局路径的相对关系,利用五次多项式生成多条待选局部路径;采用障碍物势场法模糊处理包装车间障碍的外形轮廓,并设置膨胀区间;结合待选路径采样点经过障碍物膨胀区间得到的碰撞值,以及采样点与全局规划路径的偏差值,综合评估路径的避障能力,筛选出最优的避障路径。结果 该算法能根据障碍物的大小,动态解算出与全局路径偏移量小且运动连贯的局部避障路径。结论 所提算法使得车辆完成运输任务时的效果更好,且效率更高。     

Intelligent Firefly Algorithm Deep Transfer Learning Based COVID-19 Monitoring System

With the increasing and rapid growth rate of COVID-19 cases, the healthcare scheme of several developed countries have reached the point of collapse. An important and critical steps in fighting against COVID-19 is powerful screening of diseased patients, in such a way that positive patient can be treated and isolated. A chest radiology image-based diagnosis scheme might have several benefits over traditional approach. The accomplishment of artificial intelligence (AI) based techniques in automated diagnoses in the healthcare sector and rapid increase in COVID-19 cases have demanded the requirement of AI based automated diagnosis and recognition systems. This study develops an Intelligent Firefly Algorithm Deep Transfer Learning Based COVID-19 Monitoring System (IFFA-DTLMS). The proposed IFFA-DTLMS model majorly aims at identifying and categorizing the occurrence of COVID19 on chest radiographs. To attain this, the presented IFFA-DTLMS model primarily applies densely connected networks (DenseNet121) model to generate a collection of feature vectors. In addition, the firefly algorithm (FFA) is applied for the hyper parameter optimization of DenseNet121 model. Moreover, autoencoder-long short term memory (AE-LSTM) model is exploited for the classification and identification of COVID19. For ensuring the enhanced performance of the IFFA-DTLMS model, a wide-ranging experiments were performed and the results are reviewed under distinctive aspects. The experimental value reports the betterment of IFFA-DTLMS model over recent approaches.     

Effects of bottom profile on the circulation and classification of particles in cylindrical hydrocyclones

The cylindrical hydrocyclone has been increasingly used in coarse classification due to its reduced fine particle entrainment, but the loss of coarse particles to overflow remains an intractable problem. Based on the notion that the strong circulation flow caused by the flat bottom structure bears primary responsibility for the problem, this study designs eight unique bottom profiles to regulate the particle circulating flow and attempts to correlate particle circulation flow with classification performance. The effects of the bottom profile on flow field characteristics, particle spatial distribution, circulation flow rates, and grade efficiency are explored in detail using validated models in a Φ200 mm cylindrical hydrocyclone. The findings suggest that bottom profiles have the greatest effect on the axial velocity near the bottom and the grade efficiency of intermediate and coarse particles, while all unique designs have the potential to lower turbulence intensity. An ascending segment near the wall or a descending segment near the axis can help to mitigate the misplacement of coarse particles by reducing particle circulation flow without affecting the entrainment of fines appreciably. Additionally, two circles are found on each side of the cut plane, which is conducive to releasing coarse particles from the circulation flow. Regulation of particle circulation flow by adjusting bottom profile parameters can improve separation performance.     

A Hemispherical Image Sensor Array Fabricated with Organic Photomemory Transistors

Hemispherical image sensors simplify lens designs, reduce optical aberrations, and improve image resolution for compact wide-field-of-view cameras. To achieve hemispherical image sensors, organic materials are promising candidates due to the following advantages: tunability of optoelectronic/spectral response and low-temperature low-cost processes. Here, a photolithographic process is developed to prepare a hemispherical image sensor array using organic thin film photomemory transistors with a density of 308 pixels per square centimeter. This design includes only one photomemory transistor as a single active pixel, in contrast to the conventional pixel architecture, consisting of select/readout/reset transistors and a photodiode. The organic photomemory transistor, comprising light-sensitive organic semiconductor and charge-trapping dielectric, is able to achieve a linear photoresponse (light intensity range, from 1 to 50 W m⁻²), along with a responsivity as high as 1.6 A W⁻¹ (wavelength = 465 nm) for a dark current of 0.24 A m⁻² (drain voltage = −1.5 V). These observed values represent the best responsivity for similar dark currents among all the reported hemispherical image sensor arrays to date. A transfer method was further developed that does not damage organic materials for hemispherical organic photomemory transistor arrays. These developed techniques are scalable and are amenable for other high-resolution 3D organic semiconductor devices.     

A node-based smoothed finite element method (NS-FEM) for upper bound solution to visco-elastoplastic analyses of solids using triangular and tetrahedral meshes

A node-based smoothed finite element method (NS-FEM) was recently proposed for the solid mechanics problems. In the NS-FEM, the system stiffness matrix is computed using the smoothed strains over the smoothing domains associated with nodes of element mesh. In this paper, the NS-FEM is further extended to more complicated visco-elastoplastic analyses of 2D and 3D solids using triangular and tetrahedral meshes, respectively. The material behavior includes perfect visco-elastoplasticity and visco-elastoplasticity with isotropic hardening and linear kinematic hardening. A dual formulation for the NS-FEM with displacements and stresses as the main variables is performed. The von-Mises yield function and the Prandtl–Reuss flow rule are used. In the numerical procedure, however, the stress variables are eliminated and the problem becomes only displacement-dependent. The numerical results show that the NS-FEM has higher computational cost than the FEM. However the NS-FEM is much more accurate than the FEM, and hence the NS-FEM is more efficient than the FEM. It is also observed from the numerical results that the NS-FEM possesses the upper bound property which is very meaningful for the visco-elastoplastic analyses which almost have not got the analytical solutions. This suggests that we can use two models, NS-FEM and FEM, to bound the solution, and can even estimate the global relative error of numerical solutions.     

Resource service optimal-selection based on intuitionistic fuzzy set and non-functionality QoS in manufacturing grid system

In manufacturing grid (MGrid) system, according to functional requirements of a task, there exist a lot of resource services which have similar functional characteristics. Multiple resource services with similar functional characteristics raise the concern over resource service optimal-selection (RSOS). It is important to select the optimal resource service according to their non-functionality characteristics or quality of service (QoS). However, QoS attributes are not easy to measure due to their complexity and involvement of ill-structured information. In this study, user’s feeling is taken into account in RSOS in an MGrid system. The non-functionality QoS evaluation of resource services is based on users’ feeling and transaction experiences using intuitionistic fuzzy set (IFS). Furthermore, the dynamics of non-functionality QoS is considered, and a time-decay function is introduced into non-functionality QoS evaluation. A new method is proposed for RSOS based on IFS and non-functionality QoS, and the procedures are presented in detail. A practice case study is used to illustrate the proposed method and procedure. The performance and advantage of the proposed method are discussed.     

Optimization of optical lens-controlled scanning electron microscopic resolution using generalized regression neural network and genetic algorithm

A scanning electron microscope (SEM) is a sophisticated equipment employed for fine imaging of a variety of surfaces. In this study, prediction models of SEM were constructed by using a generalized regression neural network (GRNN) and genetic algorithm (GA). The SEM components examined include condenser lens 1 and 2 and objective lens (coarse and fine) referred to as CL1, CL2, OL-Coarse, and OL-Fine. For a systematic modeling of SEM resolution (R), a face-centered Box–Wilson experiment was conducted. Two sets of data were collected with or without the adjustment of magnification. Root-mean-squared prediction error of optimized GRNN models are GA 0.481 and 1.96×10^-12 for non-adjusted and adjusted data, respectively. The optimized models demonstrated a much improved prediction over statistical regression models. The optimized models were used to optimize parameters particularly under best tuned SEM environment. For the variations in CL2 and OL-Coarse, the highest R could be achieved at all conditions except a larger CL2 either at smaller or larger OL-Coarse. For the variations in CL1 and CL2, the highest R was obtained at all conditions but larger CL2 and smaller CL1.     

Energy-efficient adaptive sensor scheduling for target tracking in wireless sensor networks

Sensor scheduling is essential to collaborative target tracking in wireless sensor networks (WSNs). In the existing works for target tracking in WSNs, such as the information-driven sensor query (IDSQ), the tasking sensors are scheduled to maximize the information gain while minimizing the resource cost based on the uniform sampling intervals, ignoring the changing of the target dynamics and the specific desirable tracking goals. This paper proposes a novel energyefficient adaptive sensor scheduling approach that jointly selects tasking sensors and determines their associated sampling intervals according to the predicted tracking accuracy and tracking energy cost. At each time step, the sensors are scheduled in alternative tracking mode, namely, the fast tracking mode with smallest sampling interval or the tracking maintenance mode with larger sampling interval, according to a specified tracking error threshold. The approach employs an extended Kalman filter (EKF)-based estimation technique to predict the tracking accuracy and adopts an energy consumption model to predict the energy cost. Simulation results demonstrate that, compared to a non-adaptive sensor scheduling approach, the proposed approach can save energy cost significantly without degrading the tracking accuracy.