Recognition of machining features - a hybrid approach

This paper describes a hybrid system which endeavours to recognize machining features automatically from a boundary representation (b-rep)-based solid modeller. The graph-based approach and the volume approach are adopted in consecutive stages in a prototype feature recognition system to combine the positive aspects of both strategies. The graph-based approach is based on feature edge sequence (FES) graph, a new graph structure introduced in this system. The FES graph approach is used to extract primitive features from the three-dimensional solid model; and the volume decomposition approach is incorporated to generate multiple interpretations of the feature sets. In addition, a neural network (NN)-based technique is used to tackle the problem of nonorthogonal and arbitrary features. Using the hybrid system, a workpiece designed in b-rep solid modeller will be interpreted and represented by a set of primitive features attached with significant manufacturing parameters, including multiple interpretations, tool directions and machining sequences, etc. The overall hybrid system is able to transform a pure geometric model into a machining feature-based model which is directly applicable for downstream manufacturing applications.     

Recognition and interpretation of interacting and non-interacting features using spatial decomposition and Hamiltonian path search

Since it is a complex task to formalize the feature recognition problem explicitly, a large variety of systems has been developed. One of the problems these systems have to overcome is the recognition and interpretation of interacting features. A fair success has been achieved in surface based methods to recognize certain classes of interacting features. However the problem remains for general cases of interacting features. Recently much effort has been focused on the volumetric approach. We present here the current state of a volumetric feature recognition method. The system considers interacting features in prismoidal parts and it operates in two stages: (1) recognition of regions of interest: a spatial decomposition of the space bounded by a predefined circumscribing volume is performed. A ‘cell evaluated and directed adjacency graph’ is then established. This graph is traversed to identify cavity volumes. (2) interpretation: cavity volumes made up of more than one cell can be produced by different machining operations. A graph-based decomposition method and Hamiltonian path search are combined to generate interpretations which correspond to optimal machining. The system CEDAG developed in this work uses a cell-face directed graph and contrasts the face-edge and edge-vertex graphs encountered in most conventional graph-based recognition methods.     

Neural-network classifiers for automatic real-world aerial image recognition

We describe the application of the multilayer perceptron (MLP) network and a version of the adaptive resonance theory version 2-A (ART 2-A) network to the problem of automatic aerial image recognition (AAIR). The classification of aerial images, independent of their positions and orientations, is required for automatic tracking and target recognition. Invariance is achieved by the use of different invariant feature spaces in combination with supervised and unsupervised neural networks. The performance of neural-network-based classifiers in conjunction with several types of invariant AAIR global features, such as the Fourier-transform space, Zernike moments, central moments, and polar transforms, are examined. The advantages of this approach are discussed. The performance of the MLP network is compared with that of a classical correlator. The MLP neural-network correlator outperformed the binary phase-only filter (BPOF) correlator. It was found that the ART 2-A distinguished itself with its speed and its low number of required training vectors. However, only the MLP classifier was able to deal with a combination of shift and rotation geometric distortions.     

Feature patterns in recognizing non-interacting and interacting primitive,circular and slanting features using a neural network

Many varied techniques have long been suggested for the recognition of features from solid modellers, and the systems which have incorporated these techniques have achieved a moderate success. However the problem of recognition of the wide variety of features, e.g. interacting and non-interacting primitive, circular and slanting features, that any real life component may have, requires complex systems which are inflexible and hence limited in their use. Here, we present a simple and flexible system in which the features are defined as patterns of edges and vertices to deal with all the above types of features. The system starts by searching a B-rep solid model, using a cross-sectional layer method, for volumes which can be considered to represent features. Once the volumes are detected, their edges and vertices are processed and arranged into feature patterns which are used as input for a neural network to recognize the features. Simple conventions used in this work enable the creation of feature patterns for primitive, circular and slanting features. Learning, generalizing and tolerating incomplete data are some of the neural network's attributes exploited in this work to deal with interacting and non-interacting features.     

基于随机森林的统计控制图模式识别研究

引入随机森林方法进行统计控制图模式识别的研究。提取了控制图的统计特征和形状特征,设计了5种不同的特征组合方法,利用蒙特卡洛仿真方法产生训练数据集和测试数据集,选取了常用的3种模式识别方法（支持向量机方法、人工神经网络方法、决策树方法）进行对比。实验结果表明,随机森林方法相比其他3种分类器方法,在分类准确率和消耗时间两个维度上都有明显优势,可以应用于统计过程控制图模式识别。     

Sequence-dependent clustering of parts and machines: A Fuzzy ART neural network approach

This study addresses the problem of identifying families of parts having a similar sequence of operations. This is a prerequisite for the implementation of cellular manufacturing, group technology, just-in-time manufacturing systems, and for streamlining material flows in general. A pattern recognition approach based on artificial neural networks is proposed, and it is shown that the Fuzzy ART neural network can be effectively utilized for this application. First, a representation scheme for operation sequences is developed, followed by an illustrative example. A more comprehensive experimental verification, based on the mixture-model approach is then performed to evaluate its performance. The experimental factors include size of the part-machine matrix, proportion of voids, proportion of exceptional elements, and vigilance threshold. It is shown that this neural network is effective in identifying good clustering solutions, consistently and with relatively fast execution times.     

用小波神经网络检测结构损伤

用小波和神经网络ART2相结合的方法检测结构的损伤位置。给出了小波变换和人工神经网络的基本理论及其用于损伤检测的原理与特点。通过把小波变换作为神经网络的前处理来构造小波神经网络。首先通过数值试验检验了小波消噪和小波神经网络损伤检测的能力。然后在一个框架结构模型上进行了试验。实验证明这种方法使网络抗噪声能力增强,使损伤识别的效果更好。ART2网络具有自动从环境中学习的能力,能自动的给出新的识别输出。     

Tool Wear Monitoring in Drilling Using Multiple Feature Fusion of the Cutting Force

1. IntroductionTool wear brings about vibration of the machine tool and deterioration of surface roughness and dimensional accuracy of the workpiece, and even causes tool damage and workpiece damage as well as downtime of the machine tool when the tool is badly worn. It is reported that 75% of all the downtime of equipment is owed to tool failure in production. So on-line monitoring of tool condition and tool wear is a key technique that realizes high efficiency and automation of the machining…     

Recognition of interacting rotational and prismatic machining features from 3-D mill-turn parts

Current feature recognition methods generally recognize and classify machining features into two classes: rotational features and prismatic features. Based on the different characteristics of geometric shapes and machining methods, rotational features and prismatic features are recognized using different methods. Typically, rotational features are recognized using two-dimensional (2-D) edge and profile patterns. Prismatic features are recognized using 3-D geometric characteristics, for example, patterns in solid models such as 3-D face adjacency relationships. However, the current existing feature recognition methods cannot be applied directly to a class of so-called mill-turn parts where interactions between rotational and prismatic features exist. This paper extends the feature recognition domain to include this class of parts with interacting rotational and prismatic features. A new approach, called the machining volume generation method, is developed. The feature volumes are generated by sweeping boundary faces along a direction determined by the type of machining operation. Different types of machining features can be recognized by generating different forms of machining volumes using various machining operations. The generated machining volumes are then classified using face adjacency relationships of the bounding faces. The algorithms are executed in four steps, classification of faces, determining machining zones, generation of rotational machining volumes and prismatic machining volumes, and classification of features. The algorithms are implemented using the 3-D boundary representation data modelled on the ACIS solid modeller. Example parts are used to demonstrate the developed feature recognition method.     

Identification of machining features based on available resources of cutting tools

In research on machining feature recognition, the problems of interacting features and availability of cutting tools are considered two major obstacles for developing industrial applications. In this research, a new machining feature recognition approach is developed to address these problems. In this work, a new concept called cutting mode is introduced to associate generic machining surfaces and cutting motions. In the feature recognition process, the machining surfaces of a part are first mapped to cutting modes, and these cutting modes are further mapped to available cutting tools. Among all the created candidate machining processes, heuristic rules are employed to identify the optimal solution that requires the minimum number of setups. When a number of machining surfaces are associated with a cutting tool in the same setup, these surfaces are grouped as a machining feature. Therefore the interacting features are recognised by the different cutting tools to produce these features. A database of available cutting tools is used to avoid the identification of features which cannot be machined in a machine shop. Three mechanical parts with interacting features are selected in the case studies to demonstrate the effectiveness of the developed approach.     

Evolutionary feature selection applied to artificial neural networks for wood-veneer classification

This paper presents the application of FeaSANNT, an evolutionary algorithm for optimization of artificial neural networks, to the training of a multi-layer perceptron for identification of defects in wood veneer. Given a fixed artificial neural network structure, FeaSANNT concurrently evolves the input feature vector and the network weights. The novelty of the method lies in the implementation of the embedded approach in an evolutionary feature selection paradigm. Experimental tests show that the proposed algorithm produces high-performing solutions with robust learning results. A significant reduction of the set of veneer features is obtained. Experimental comparisons are made with a previous method based on statistical filtering of the input features and a standard genetic wrapper algorithm. In the first case, FeaSANNT greatly reduces the feature set with no degradation of the neural network accuracy. Moreover, FeaSANNT entails lower design costs, since feature selection is fully automated. In the second case, the proposed algorithm achieves superior results in terms of identification accuracy and reduction of the feature set. FeaSANNT involves also lower computational costs than the standard evolutionary wrapper approach and eases the algorithm design effort. Limited overlapping is observed between the patterns of features selected by the three algorithms. This result suggests that the full feature set contains mainly redundant attributes.     

A Fuzzy Neural Network for Fault Pattern Recognition

1　ＩｎｔｒｏｄｕｃｔｉｏｎＩｎｍａｔｈｅｍａｔｉｃｓ,ｆａｕｌｔｒｅｃｏｇｎｉｔｉｏｎｃａｎｂｅｓｕｍｍｅｄｕｐａｓａｍａｐｐｉｎｇｐｒｏｂｌｅｍｂｅｔｗｅｅｎｆａｕｌｔａｇｇｒｅ ｇａｔｅａｎｄｃｈａｒａｃｔｅｒａｇｇｒｅｇａｔｅ .Ｔｈｅｍａｐｐｉｎｇｂｅｔｗｅｅｎａｇｇｒｅｇａｔｅｓｉｓｃａｌｌｅｄａｍａｐｐｉｎｇｆｕｎｃｔｉｏｎ ;ｋｉｎｄｓｏｆｍａｐｐｉｎｇｆｕｎｃｔｉｏｎｓｃａｎｂｅｆｏｒｍｅｄｆｏｒｆａｕｌｔｐａｔｔｅｒｎｒｅｃｏｇｎｉｔｉｏｎ .Ｔｈｅｔｒａｄｉｔｉｏｎａｌｐａｔｔｅｒ…     

基于深度置信网络与信息融合的齿轮故障诊断方法

针对齿轮在复杂运行工况下故障特征提取困难,传统故障诊断方法的识别精度易受人工提取特征的影响,以及单传感器获取信息不全面等问题,提出基于深度置信网络(DBN)与信息融合的齿轮故障诊断方法。通过多传感器信息融合技术对每个传感器采集的振动信号进行数据层融合;利用DBN进行自适应特征提取从而实现故障分类。为了避免因人为选择DBN结构参数,导致模型识别精度下降的问题,利用改进的混合蛙跳算法(ISFLA)对DBN结构参数进行优化。试验表明,与BP神经网络、未经优化的DBN以及单传感器故障诊断相比,该研究提出的信息融合及优化方法具有更高的故障识别精度。     

A combinatorial optimisation approach for recognising interacting machining features in mill-turn parts

Feature interactions may result in many process alternatives in part machining. Traditional process planning methods only identify one of the options, which is usually not optimal in the sense of engineering. This paper presents an optimisation approach to handle the interacting feature recognition problem in mill-turn parts. The approach subdivides the material removal volume into cells first and then it combines the cells into features. Here, a two-level cell combination method is developed. On the lower level, individual features are formed by searching the combinations of cells near a given part face; on the upper level, the feature distributions are explored by rearranging the order of part faces for feature formation. In order to optimise the feature distribution, a novel optimisation model is proposed, which quantitatively distinguishes its options by considering the factors of feature numbers, tool approaching directions, cutting directions and surface roughness. The combinatorial optimisation problem is solved with the simulated annealing algorithm. Instead of searching cell combinations directly, the proposed method explores different part face sequences, which drastically reduces the search space. The case studies show that the proposed approach can effectively handle the traditional difficulty in recognising the interacting features for mill-turn parts.     

神经网络在二维零件图特征识别中的应用

研究用感知器和三层ＢＰ神经网络识别二维零件图的形状特征信息。提出了零件图扩展属性矩阵的概念及求法,对ＡutoCAD R14进行了二 开发,完善了笔者开发的二维回转类零件图特征识别和提取系统。     

发动机喘振故障检测的神经网络免疫识别模型

提出了一种航空发动机喘振故障检测的神经网络免疫识别模型。该模型利用人工免疫系统的反面选择原理来构建神经网络检测器,通过训练将失速压力信号的模式特征存储在分布的检测器中。检测器用于捕获信号的失速模式特征,当检测器与特征样本匹配时则激活该检测器,根据检测器的激活情况来发现失速点。对某型涡喷发动机压力测量信号的分析结果表明,该方法对由失速气团造成的压力信号突变具有较强的分辨力,可以用于发动机喘振的早期检测。     

基于特征融合进行活动识别的DCNN 5法

研究了输入是可穿戴传感器获得的多通道时间序列信号,输出是预定义的活动的活动识别模型,指出活动中的有效特征的提取目前多依赖于手工和浅层特征学习结构,不仅复杂而且会导致识别准确率下降;基于深度学习的卷积神经网络（ CNN）不是对时间序列信号进行手工特征提取,而是自动学习最优特征;目前使用卷积神经网络处理有限标签数据仍存在过拟合问题。因此提出了一种基于融合特征的系统性的特征学习方法用于活动识别,用ImageNet16对原始数据集进行预训练,将得到的数据与原始数据进行融合,并将融合数据和对应的标签送入有监督的深度卷积神经网络（ DCNN ）中,训练新的系统。在该系统中,特征学习和分类是相互加强的,它不仅能处理端到端的有限数据问题,也能使学习到的特征有更强的辨别力。与其他方法相比,该方法整体精度从87.0％提高到87.4％。     

嵌入注意力机制的卷积神经网络水声目标识别

针对低信噪比水声目标单一特征识别率低,稳健性差的问题,提出一种基于注意力机制和多尺度残差卷积神经网络（Multi-scale Residual CNN with Attention,MR-CNN-A）进行特征融合的识别方法。该方法根据多尺度卷积核与特征图形成多分辨率分析关系,并以此通过注意力机制实现优势特征权值提取与融合,从而提高模型在文中水声数据集上提取目标噪声特征和分类识别的稳健性与抗噪能力。开展了4类舰船噪声和海洋环境噪声的识别试验、水下和水面自主式水下航行器的识别试验,以及不同信噪比条件下目标噪声的识别试验。结果表明：对于文中所涉及的水声目标噪声和人工高斯白噪声干扰,该网络模型识别正确率明显高于支持矢量机与简单卷积神经网络,且对高斯白噪声的抑制能力远强于支持矢量机与简单卷积神经网络,稳健性好,模型复杂度小。     

基于自身注意力时空特征的语音情感识别算法

针对语音情感识别中无法对关键的时空依赖关系进行建模,导致识别率低的问题,提出一种基于自身注意力（self-attention）时空特征的语音情感识别算法,利用双线性卷积神经网络、长短期记忆网络和多组注意力（multi-headattention）机制去自动学习语音信号的最佳时空表征。首先提取语音信号的对数梅尔（log-Mel）特征、一阶差分和二阶差分特征合成3D log-Mel特征集作为卷积神经网络的输入;然后综合考虑空间特征和时间依赖性关系,将双线性池化和双向长短期记忆网络的输出融合得到空间-时间特征表征,利用多组注意力机制捕获判别性强的特征;最后利用softmax函数进行分类。在IEMOCAP和EMO-DB数据库上进行实验,结果表明两种数据库的识别率分别为63.12%和87.09%,证明了此方法的有效性。     

基于卷积神经网络和Transformer网络的鸟声识别

针对传统鸟声识别算法中特征提取方式单一、分类识别准确率低等问题,提出一种结合卷积神经网络和Transformer网络的鸟声识别方法。该方法综合考虑网络局部特征学习和全局上下文依赖性构造,从原始鸟声音频信号中提取短时傅里叶变换(Short Time Fourier Transform,STFT)语谱图特征,将其输入到卷积神经网络(ConvolutionalNeural Network,CNN)中提取局部频谱特征信息,同时提取鸟声信号的对数梅尔特征及一阶差分、二阶差分特征用于合成梅尔频率倒谱系数(Mel Frequency Cepstrum Coefficient,MFCC)混合特征向量,将其输入到Transformer网络中获取全局序列特征信息,最后融合所提取的特征可得到更丰富的鸟声特征参数,通过Softmax分类器得到鸟声识别结果。在Birdsdata和xeno-canto鸟声数据集上进行实验,平均识别准确率分别达到了97.81%和89.47%。实验结果表明该方法相较于其他现有的鸟声识别模型具有更高的识别准确率。