High-energy spectrophotometer for rapid constituent analysis in the range of 0.25-2.4 mum

This publication describes a constituent analyzer which uses a concave holographic grating for high-energy throughput, a conjugate cam drive to allow up to ten spectral scans per second, a microprocessor for data gathering and system control, and a minicomputer for the complex statistical computations required in the interpretation of the spectral data gathered. The instrument was specifically developed for quantitative measurement of food constituents in food processing and quality control. However, as a general state-of-the-art spectrophotometer it was found to be a valuable tool in other disciplines such as industrial, clinical, and medical applications.     

基于多幅X射线数字图像的缺陷自动识别技术

在现有的X射线数字图像自动识别方法中,多采用对单幅数字图像进行孤立评判的方法。由于此类方法中阈值选取难以最优化,因而存在一定的误判率。为了解决这一问题,提出了一种X射线数字图像自动识别新方法。该方法将识别过程分为两步:缺陷提取和缺陷跟踪。第一步利用传统方法在每幅图像中分离出潜在缺陷。这一步保证真缺陷能全部提取出来,而不考虑伪缺陷的数量。第二步力图找出同一试件不同图像中分离出的缺陷之间的相互关系。如果第一步某一图像中分离出的某一缺陷在其他图像中都找不到相对应的缺陷区域,就定义该缺陷为伪缺陷,也就是说,真缺陷在不同图像中必须满足一定的几何关系。多幅图像中的缺陷跟踪综合利用了极线约束、三维重建和三线性约束等立体视觉算法。该方法的检测效果已经利用航空发动机叶片X射线数字图像得到验证。试验结果表明:利用该方法可以提高真缺陷的识别率,降低误判率。     

In situ monitoring of selective laser melting using plume and spatter signatures by deep belief networks

Critical quality issues such as high porosity, cracks, and delamination are common in current selective laser melting (SLM) manufactured components. This study provides a flexible and integrated method for in situ process monitoring and melted state recognition during the SLM process, and it is useful for process optimization to decrease part quality issues. The part qualities are captured by images obtained from an off-axis setup with a near-infrared (NIR) camera. Plume and spatter signatures are closely related to the melted states and laser energy density, and they are employed for the SLM process monitoring in an adapted deep belief network (DBN) framework. The melted state recognition with the improved DBN and original NIR images requires little signal preprocessing, less parameter selection and feature extraction, obtaining the classification rate 83.40% for five melted states. Compared to the other methods of neural network (NN) and convolutional neural networks (CNN), the proposed DBN approach is identified to be accurate, convenient, and suitable for the SLM process monitoring and part quality recognition.     

Development of a framework for pick-and-pass order picking system design

In today??s competitive scenario of increasingly faster deliveries and smaller order sizes, a solution that is being used more frequently is the pick-and-pass Order Picking System (OPS). The purpose of this study is to define a framework for the pick-and-pass system design, by expanding previous literature on this topic. The framework aims at minimising the overall picking costs while respecting the required service level (i.e. order throughput time), and can be easily applied to the selection stage of OPS design. The number of zones and the number of pickers per zone are among the main project data for the framework application. Analytical models are used to estimate the travel distance, and network queuing theory to analyse the mean order throughput. On the basis of the proposed framework, pick-and-pass system performance is examined as a function of two typical design conditions: order size and number of items.     

Extraction and parametrization of grain boundary networks in glacier ice,using a dedicated method of automatic image analysis

Microstructure analysis of polar ice cores is vital to understand the processes controlling the flow of polar ice on the microscale. This paper presents an automatic image processing framework for extraction and parametrization of grain boundary networks from images of the NEEM deep ice core. As cross‐section images are acquired using controlled surface sublimation, grain boundaries and air inclusions appear dark, whereas the inside of grains appears grey. The initial segmentation step of the software is to separate possible boundaries of grains and air inclusions from background. A Machine learning approach is utilized to gain automatic, reliable classification, which is required for processing large data sets along deep ice cores. The second step is to compose the perimeter of section profiles of grains by planar sections of the grain surface between triple points. Ultimately, grain areas, grain boundaries and triple junctions of the later are diversely parametrized. High resolution is achieved, so that small grain sizes and local curvatures of grain boundaries can systematically be investigated.     

A robust quantized consensus protocol for discrete-time multi-agent systems with additive noise

This paper investigates the consensus problem of multiple discrete-time integrator agents with communication constraints and additive process noise. It proposes a protocol to achieve the approximate consensus of agents over inter-agent communication networks with finite bit rates. Under that protocol, dynamic encoding and decoding algorithms are implemented for each pair of neighboring agents to transmit quantized states at a finite bit rate. With received quantized states of neighboring agents, the control input of each agent is locally computed. Particularly input saturation is introduced into the local controllers of agents and places both lower and upper bounds on the local control inputs of agents. These control input bounds can be known in advance and greatly enhance the robustness of the consensus protocol. Under the proposed protocol, the approximate consensus can be guaranteed at any finite bit rate to encode the states of agents. It is shown that even a single bit per time step is enough for the desired approximate consensus. The additive process noise does not increase the bit rate required for that approximate consensus. Moreover, the proposed consensus protocol can be designed with only an upper bound on the number of agents and is more robust than some previous consensus protocols which may require the global information of the inter-agent network topology, such as the second largest eigenvalue of the Laplacian matrix. Even when some communication links are broken due to communication failure or some nodes leave, the same set of consensus parameters can still robustly guarantee the expected approximate consensus. Simulations are conducted to illustrate the effectiveness of the proposed quantized consensus protocol.     

一类异类无线传感器网络节点调度问题研究

针对一类以配置了多种传感器的节点组成的,部分传感器完全覆盖,部分传感器局部覆盖的异类无线传感器网络节点调度问题,提出了一种基于改进遗传算法的优化策略.在构建网络模型的基础上,建立了节点调度分化策略,提出了冗余信息度的概念来描述网络能耗效率,并设计了以冗余信息度和不同传感器目标区域感知覆盖率为优化目标的改进多目标遗传算法NSGAⅡ,用于求解节点分化策略.仿真结果表明,该方法可以通过迭代得到收敛的Pareto最优解,并为传感器网络提供一个多目标Pareto最优节点分化策略方案集,供不同应用选择.     

多层网络中基于资源优化的配置方式

网络生存性是衡量网络性能优劣的重要指标,其与实际网络应用关系最为密切,网络生存性的优劣不仅体现在所应用的生存策略上,资源的优化配置也是影响网络生存性的关键因素。在多层网络资源优化方法中,网络中空闲资源配置方式至关重要,是网络发展迫切需要解决的关键问题之一。首先分析了多层网络资源配置方法的研究现状,其次介绍了多层网络中所采用的网络模型,结合网络模型给出了2种网络资源的配置方式,并对这2种方式进行了比较。     

Flat field correction for high‐throughput imaging of fluorescent samples

Vignetting of microscopic images impacts both the visual impression of the images and any image analysis applied to it. Especially in high‐throughput screening high demands are made on an automated image analysis. In our work we focused on fluorescent samples and found that two profiles (background and foreground) for each imaging channel need to be estimated to achieve a sufficiently flat image after correction. We have developed a method which runs completely unsupervised on a wide range of assays. By adding a reliable internal quality control we mitigate the risk of introducing artefacts into sample images through correction. The method requires hundreds of images for the foreground profile, thus limiting its application to high‐throughput screening where this requirement is fulfilled in routine operation.     

Flow imaging for multi-component flow measurement

For pseudo-homogeneous flows, measurements of density and mean velocity can give the component mass flow rate of a two-component mixture. However, for accurate measurement of non-homogeneous flow rate, the density and velocity distribution across the cross-section of the pipe must be known. The most practical way of obtaining this information is by using the flow imaging technique.

A recently developed capacitance system gives 60 frames per second images of oil/water flow in a 78 mm pipe. The target spatial resolution is one part in 20 by distance (one in 400 by area). The electrical properties of each imaged boundary are functionally related to the imaged value, so the component ratio of a two-component mixture within a boundary can be measured, although individual particles cannot be imaged. Design data shows how the basic system can be part of a complete system for component mass flow measurement.     

Measurement and evaluation of surface roughness based on optic system using image processing and artificial neural network

The traditional devices, used to measure the surface roughness, are very sensitive, and they are obtained by scratching the surface of materials. Therefore, the optic systems are used as alternatives to these devices to avoid the unwanted processes that damage the surface. In this study, face milling process was applied to American Iron and Steel Institute (AISI) 1040 carbon steel and aluminium alloy 5083 materials using the different tools, cutting speeds and depth of cuts. After these processes, surface roughness values were obtained by the surface roughness tester, and the machined surface images were taken using a polarise microscope. The obtained images were converted into binary images, and the images were used as input data to train network using the MATLAB neural network toolbox. For the training networks, log-sigmoid function was selected as transfer function, scaled conjugate gradient (SCG) algorithm was used as training algorithm, and performance of the trained networks was achieved as an average of 99.926 % for aluminium alloy (AA) 5083 aluminium and as an average of 99.932 % for AISI 1040 steel. At the end of the study, a prediction programme for optical surface roughness values using MATLAB m-file and GUI programming was developed. Then, the prediction programme and neural network performance were tested by the trial experiments. After the trial experiments, surface roughness values obtained with stylus technique for the carbon steel and aluminium alloy materials were compared with the developed programme values. When the developed programme values were compared with the experimental results, the results were confirmed each other at a rate of 99.999 %.     

集成神经网络在设备诊断中的应用

本文从诊断工程实际出发 ,从信息融合的角度提出了集成神经网络的概念 ,并讨论了实现的策略。诊断实例表明 ,集成神经网络的建立 ,充分利用了各种信息 ,可有效提高确诊率     

Emulating concurrency in a circuit card assembly system

Typical component-placement systems for populating surface mount technology printed circuit boards now exhibit a high degree of concurrency in their functional operations. This concurrency ideally yields high burst-rate estimates of throughput. However, if the concurrency is not properly understood and exploited, the burst rate is severely degraded, as exhibited by process rates observed in the actual production environment. This discernment requires an experimental characterization of the system's functional operation, which must also reflect the peculiarities of the controller. Such an experimental analysis is an essential precursor to performance-optimization procedures of numerically controlled flexible manufacturing systems. This article describes our analysis of an extremely complex workcell with a high degree of concurrency. Due to its enveloping complexity, the methodological framework for the analysis should be applicable to a broad class of concurrent systems. Empirically verifying the characterization required the development of an emulator that quantitatively defines the system to be modeled. As such, it is a numerical, off-line design and analysis tool. It has been utilized to obtain the process rate for particular products, preevaluate proposed engineering changes, interactively construct setups and sequences, and obtain parameters required for line-balancing procedures.     

Concurrent design of manufacturing cell and control functions: A neural network approach

This paper presents a neural network based decision support system (DSS) for use in concurrently determining cell configuration, operation plans, and complexity requirements of cell control functions. Advanced simulators and neural network technology are used in developing the DSS. Simulation experiments were conducted with many possible combinations of design changes to generate training pairs for a neural network. Complexity of cell control functions required by each design option was assessed, based on operational requirements, and was used to train another neural net. Once both neural networks are properly trained, one network can be used to predict the cell design configuration given a set of desirable cell performance measures, while the other network can be used to identify complexity requirements of the cell control functions by using the output provided by the first network as input to the second neural net. An operation-driven cell design methodology was applied to sequentially predict requirements of both cell configuration and cell control functions from the trained neural networks. This innovative new design methodology was illustrated via a successful implementation exercise in acquiring a real automated manufacturing cell at industrial settings. The exercise proves that such a DSS serves well as an effective tool for cell designers and the management in determining appropriate cell configuration and cell control functions at the design stage.     

Spot counting to locate fetal cells in maternal blood and tissue: a comparison of manual and automated microscopy

BACKGROUND: Fetal cell detection in maternal tissue requires an accurate, efficient, and reproducible microscopy method. Our objective was to compare manual scoring to a commercially available automated scanning system for the detection of chromosome signals by fluorescence in situ hybridization (FISH). METHODS: X and Y chromosome FISH signals were detected on slides of calibrated mixtures of blood, paraffin-embedded liver sections, and post-termination blood. For manual scoring (400x magnification), the number of cells located and duration of scoring were recorded. For automated scanning using the Metasystems Metafer3/Metafer4 Scanning System (200x magnification), duration of scanning, number of gallery images generated, duration of manual review of gallery images, and number of confirmed fetal cells were recorded. RESULTS: From all slides the number of target fetal cells located by manual and automated microscopy was highly correlated (r = 0.90). However, automated scanning required on average 4-fold more time than manual scoring (P < 0.0001), with an average automated scanning time of 9.7 h per slide compared with 2.4 h per slide when scored manually. CONCLUSIONS: In general, the accuracy of automated and manual microscopy is comparable, although manual scoring is more efficient because of the level of magnification necessary for automated scanning of cells, and a large number of gallery images generated by automated scanning that must then be reviewed manually. This suggests that when rapid analysis is required (i.e., clinical situations), manual microscopy is preferable. In contrast, automated scanning may have advantages over manual microscopy when time constraints are less imposed (i.e., research situations).     

An optimal concurrent design and loading strategy for a flexible assembly line system: to control the bottleneck

In this paper, an optimal strategy is presented at the design stage for selection of the required local storages of the workstations and the transporter stations of a finite capacity flexible assembly line system, such that the throughput rate from the system is maximised while controlling the bottleneck problem. For this purpose, a mixed non-Markovian queueing network model is presented to model its performance, a stochastic optimisation model is provided to maximise its throughput rate, and a heuristic algorithm is developed for solving it. Finally, an example is presented and the approximation results are compared against those from a simulation study.     

Practical neural networks for industrial inspection: A case study using web-process inspection

Neural networks have been successful at pattern recognition and discovery of hidden relationships amongst parameters and as such are likely supplements to the sensory systems employed in industrial applications. This paper examines four resulting issues imposed upon any industrial inspection system using a neural network: the feature set which the sensory system must provide, the accuracy of neural-network-based inspection, the robustness required of the sensory system for accurate inspection, and the computational burden imposed by accuracy requirements. This is accomplished in the context of web-process inspection, which requires rapid examination of vast amounts of data for on-line detection of faults in the sheet material. Each of the four crucial issues is addressed:

1. (i) Feature vectors with nine or 17 dimensions, created by a simulated segmented photodetector using measurement of the angular distribution over a 25° cone angle of the scattering were evaluated for inspection CrO₂-coated sheet steel samples. The scattered coherent light from the surface of the material being processed could be directly conditioned by a photodetector so as to produce this small set of features which are then examined by a neural network trained to find and categorize unsatisfactory surface conditions. Details are presented to show how a modified feature set was developed and tested after an examination of feature space. This new, smaller set proved to be more accurate than the larger set.

2. (ii) Classification by fault or no fault categorized 133 samples correctly out of 135, while there were seven errors in one attempt at classification into the various common surface faults out of the same number of test samples and nine in another. It is shown that a bit of insight in feature selection can improve the capability of the network to recognize faults.

3. (iii) The robustness issue is important since the inspection system must function in the industrial environment, where maintaining an exact alignment of the optics is not feasible. Tests are described where it is shown that fault classification using the proposed system is reasonably robust to slight variation of the angle between the laser beam and camera.

4. (iv) The computational issue is discussed in the context of the data handling requirements of the inspection system.

Deep learning model integrating features and novel classifiers fusion for brain tumor segmentation

Automatic and precise segmentation and classification of tumor area in medical images is still a challenging task in medical research. Most of the conventional neural network based models usefully connected or convolutional neural networks to perform segmentation and classification. In this research, we present deep learning models using long short term memory (LSTM) and convolutional neural networks (ConvNet) for accurate brain tumor delineation from benchmark medical images. The two different models, that is, ConvNet and LSTM networks are trained using the same data set and combined to form an ensemble to improve the results. We used publicly available MICCAI BRATS 2015 brain cancer data set consisting of MRI images of four modalities T1, T2, T1c, and FLAIR. To enhance the quality of input images, multiple combinations of preprocessing methods such as noise removal, histogram equalization, and edge enhancement are formulated and best performer combination is applied. To cope with the class imbalance problem, class weighting is used in proposed models. The trained models are tested on validation data set taken from the same image set and results obtained from each model are reported. The individual score (accuracy) of ConvNet is found 75% whereas for LSTM based network produced 80% and ensemble fusion produced 82.29% accuracy.     

Optimization of the Cascade Feed Forward Back Propagation network for defect classification in ultrasonic images

Ultrasonic Time of Flight Diffraction (TOFD) is now a well established NDE technique finding wide applications in the industry for inspection during manufacture, pre-service and also inservice. While conventionally interpretations of UT images are done by the inspector, a need has always been felt for automated evaluation and interpretation especially when large inspection volumes are involved. Apart from enhancing the speed of inspection, automated evaluation and interpretation provides better reliability of inspection. A number of approaches based on signal analysis coupled with artificial neural networks (ANN) are being tried internationally and limited success has also been obtained. This paper focuses on the development of a semi automatic toolbox for reliable and fast flaw classification in TOFD images using ANN. TOFD images are first acquired and statistical parameters such as mean, standard deviation, energy, skewness and kurtosis are calculated for the region of interest in the images. The classification of the flawed region like Crack, Lack of Fusion, Lack of Penetration, Porosity and Slag Inclusion was materialized using different ANN approaches which made use of these statistical parameters as their input. The process of optimization of a network involves comparison of classification accuracy and the sensitivity of the selected networks. The Cascade Feed Forward Back Propagation (CFBP) network with log sigmoidal activation function proved to be the optimized network model for the data set considered in this study.     

Performance comparison of assembly systems with fixed and flexible cycle times

The traditional assembly system consists of a series of balanced workstations operating at the same rate with fixed cycle times. Recent advances in technology allow more flexible assembly systems, in which workstations operate independently and cycle times vary from job to job.This article develops an analytical model for comparing the throughputs (jobs per hour) of assembly systems with fixed and variable cycle times. The throughputs are compared on a common basis by requiring that both systems allow sufficient processing time to ensure product quality and that they have the same total times in system per job.Results indicate that an assembly system with variable cycle times can operate at a significantly higher throughput than one with fixed cycle times, provided there is sufficient buffer storage space between workstations to accommodate queueing. This benefit must be weighed against possible increased capital investment and practical considerations associated with system control.