Search

In the integrated circuit (IC) packaging, the surface defect detection of flexible printed circuit boards (FPCBs) is important to control the quality of IC. Although various computer vision (CV)-based object detection frameworks have been widely used in industrial surface defect detection scenarios, FPCB surface defect detection is still challenging due to non-salient defects and the similarities between diverse defects on FPCBs. To solve this problem, a decoupled two-stage object detection framework based on convolutional neural networks (CNNs) is proposed, wherein the localization task and the classification task are decoupled through two specific modules. Specifically, to effectively locate non-salient defects, a multi-hierarchical aggregation (MHA) block is proposed as a location feature (LF) enhancement module in the defect localization task. Meanwhile, to accurately classify similar defects, a locally non-local (LNL) block is presented as a SEF enhancement module in the defect classification task. What is more, an FPCB surface defect detection dataset (FPCB-DET) is built with corresponding defect category and defect location annotations. Evaluated on the FPCB-DET, the proposed framework achieves state-of-the-art (SOTA) accuracy to 94.15% mean average precision (mAP) compared with the existing surface defect detection networks. Soon, source code and dataset will be available at https://github.com/SCUTyzy/decoupled-two-stage-framework .

In advanced integrated circuit packing, a very important substrate is high-density flexible printed circuit (FPC) board, which is small and foldable. The detection of the circular holes on FPCs’ surfaces through machine vision is very useful to identify root causes in FPC manufacturing process. However, it is a challenge as FPC images always contain many high details. This paper proposes a circle detection method with a modified tri-class thresholding-based contour extraction to detect and measure the circular holes on FPCs’ surfaces. It first extracts object contours by using a modified tri-class thresholding and then uses two circle validation criterions to eliminate noncircles: one is based on geometrical features and the other is based on the Helmholtz principle. Moreover, an iterative circle parameter refinement strategy is proposed. Experiments are carried out on 120 FPC images and the results indicate that the modified tri-class thresholding-based contour extraction method is more efficient than Canny detector, and the proposed circle detection method shows a better performance in terms of false detection rate, computation cost, and robustness than three other popular algorithms for detecting the circles in FPC images.

Most of the existing exact formation control algorithms for multiagent systems are label-specified, i.e., the desired position for each agent in the formation is pre-specified by its label. In this paper, a self-organized control algorithm is proposed for wedge formation of a team of planar kinematic robots. The desired positions for the robots in the formation are not label-specified, but determined dynamically. By constantly forming a chain of leader-follower pairs, exact formation can be achieved by pairwise leader-follower tracking. Comprehensive simulation results are provided to evaluate the performance of the proposed control algorithm.

Texture description is a challenging problem in computer vision and pattern recognition. The task of texture classification is to classify texture into the class it belongs to, which is influenced by variations in scale, illumination, and changes in perspective. There are many texture descriptors in the literature. In this paper, we combine five texture descriptors for texture classification, which obtained better performance than the single descriptor at the price of high dimensionality. To solve this problem, we proposed a novel unsupervised feature selection method based on local structure and low-rank constraints, which can not only reduce the dimensions but also further improve the classification accuracy. To evaluate the performance of combing multiple descriptors and the proposed feature selection method, we design a variety of experiments in two typical texture datasets, namely KTH-TIPS-2a and CURET. Finally, the result shows the proposed method outperforms the state-of-the-art methods.

This paper focuses on an operation optimisation problem for a class of multi-head surface mounting machines in printed circuit board assembly lines. The problem involves five interrelated sub-problems: assigning nozzle types as well as components to heads, assigning feeders to slots and determining component pickup and placement sequences. According to the depth of making decisions, the sub-problems are first classified into two layers. Based on the classification, a two-stage mixed-integer linear programming (MILP) is developed to describe it and a two-stage problem-solving frame with a hybrid evolutionary algorithm (HEA) is proposed. In the first stage, a constructive heuristic is developed to determine the set of nozzle types assigned to each head and the total number of assembly cycles; in the second stage, constructive heuristics, an evolutionary algorithm with two evolutionary operators and a tabu search (TS) with multiple neighbourhoods are combined to solve all the sub-problems simultaneously, whe...

Effective fault classification is of great significance to isolate and eliminate faults in the smart manufacturing process, especially for chemical industry. However, incipient faults and limited labeled samples make it hard to classify faults accurately. This motivates the formulation of a semi-supervised convolutional ladder network with local and global feature fusion. In the algorithm, a convolutional ladder network is developed to capture higher-order correlations from both labeled and unlabeled samples simultaneously to overcome the problem caused by limited labeled samples, skipped connections are embedded within which to make a balance between supervised and unsupervised feature learning for further identifying faults. To improve the classification performance on incipient faults, a local and global feature fusion strategy is proposed to enhance the representation of incipient faults. Furthermore, a semi-supervised dynamic data representation strategy is introduced to jointly deal with labeled and unlabeled process samples, which enables the proposed method to handle process dynamics by characterizing temporal information of process variables. Experiments on the Tennessee Eastman process show that the proposed method is effective for process fault classification when labeled samples are limited compared to the state-of-the-art algorithms.

In order to measure the error of arbitrary contours, a high accurate visual algorithm is proposed. Firstly, a subpixel edge extraction algorithm based on local grayscale fitting is used to extract subpixel contours. And then we use the coarse-to-fine matching strategy, which uses the shape context feature of contours to perform coarse matching, and then based on iterative closest points to complete fine matching step. Finally, the neighborhood algorithm is proposed to calculate the contour error. Experiments show that the accuracy of this algorithm can reach 0.5 pixel, and this algorithm can apply to error measurement of any shape of contours.

This paper deals with a scheduling optimisation problem arising in printed circuit board (PCB) assembly. In one class of PCB assembly, light-emitting diodes are to be assembled into the placement locations on PCBs by a machine with multiple pick-and-place heads. The scheduling optimisation problem is to determine the assembly sequence of placement locations and the assignment of pick-and-place heads for locations so as to minimise the assembly time. We formulate it as a mixed integer linear programming model. To solve the problem efficiently, we classify the PCBs into two types. For the first type of PCBs, on which the locations are linearly arranged, a constructive heuristic is proposed based on the analysis of the best next location after a location is assembled. For the second type of PCBs, on which the locations are circularly arranged, a heuristic based on clustering strategy and path relinking method is proposed. Computational experiments show that the solutions obtained by the two heuristics make 2...

This paper considers a slab reallocation problem arising from operations planning in the steel industry. The problem involves reallocating steel slabs to customer orders to improve the utilisation of slabs and the level of customer satisfaction. It can be viewed as an extension of a multiple knapsack problem. We firstly formulate the problem as an integer nonlinear programming (INLP) model. With variable replacement, the INLP model is then transformed into a mixed integer linear programming (MILP) model, which can be solved to optimality by MILP optimisers for very small instances. To obtain satisfactory solutions efficiently for practical-sized instances, a heuristic algorithm based on tabu search (TS) is proposed. The algorithm employs multiple neighbourhoods including swap, insertion and ejection chain in local search, and adopts solution space decomposition to speed up computation. In the ejection chain neighbourhood, a new and more effective search method is also proposed to take advantage of the str...

Cast planning is a practical problem frequently encountered in steel industry. Its task is to group charges into batches (casts) with respect to the similarities of steel-grade and dimensions between charges, taking account of the practical technique constraints on life-span of tundish. Effective cast planning can reduce the changeover cost of charges and enhance the productivity of continuous casters. The objective under our consideration is to minimize the total dissimilarity costs between the charges in the same casts, to minimize the number of casts and the number of unselected charges. A quadratic integer programming model with multiple objectives for this problem is formulated. It is NP-complete, and so an iterated local search (ILS) algorithm is developed for the problem. In this algorithm, cyclic transfer neighborhood is adopted, in which several charges are transferred among casts simultaneously as a manner of cycle. A new kick strategy is developed with the idea of assigning charges to different casts according to the dissimilarity costs between them and the casts’ central charges identified by the current solution. Computational results using real data from an advanced iron & steel company in China indicate that the ILS algorithm provides optimal solutions for small instances, and better near-optimal solutions for larger instances compared with a linear solver, Lingo 8.0, used to solve the equivalent linear integer programming obtained by transforming the original model. Totally, 92.3% of the instances are solved to global optima by the ILS algorithm while it is possible that the other 9.7% instances also are solved to optimal, which indicates the efficiency of the algorithm. At the same time, the algorithm also provides better solutions than the ones obtained by the current system used in the company.

This paper presents a new improved Hilbert-Huang Transform(HHT) schemes for fault diagnosis, which plays especially important role in the industrial production nowadays. One of the inevitably traditional problems to handle is end effect in HHT, which is caused by Gibbs' phenomenon and other mathematical or numerical problems. In order to mitigate the end effect of HHT, a boundary extension method based on the well-known Gray Prediction Model(GM) is introduced. In this article, we adopt a new the HHT optimized with GM(2,1) model. As a second-order oscillation function model, GM(2,1) is capable of forecasting non-monotone fluctuation fault signal. Time response sequence of GM(2,1) will be obtained by estimating the expression of original signal, thus extending both ends of the original signal. Then the extensional original will be transformed by HHT method, acquiring switching signals with a longer sample radius than the quondam original. We have conducted experiments of a bearing fault diagnosis for evaluating the performance of the proposed model. According to the comparison with simulation results of GM(1,1)-HHT and HHT, it is demonstrated that GM(2,1)-HHT can significantly reduce end effects and narrows remainder range.

This paper considers the SMT placement process optimization problem to minimize the assembly time of Multi-head surface mounting machine (SMM).A mathematical model is firstly established.Considering that the component pick-and-place sequencing and the feeder assignment are the two main important factors which determine the assembly time of a device,the optimization problem is divided into two sub-problems,called feeder assignment sub-problem and the component pick-place sequencing sub-problem.Then the tabu search algorithm and the improved shuffled frog leaping algorithm with mutation are applied to optimize the two sub-problems respectively.Finally,according to the thoughts of iteration and cooperation,the two interrelated sub-problems could be cooperated with each other to improve the efficiency of optimization.To verify the efficiency of the algorithm,experiments on 10 PCBs instances are executed.Experimental results show that the algorithm could obtain satisfied quasi-optimal solutions to the mounting process,and it makes an improvement 9.55% on the hybrid genetic algorithm reported in literature.

This paper discusses matching orders against slabs problem which arises in integrated steel industry. The number of surplus slabs increases responding to varying customer needs for product customization and quantities. To deal with the surplus slabs, it is important to optimize the matching orders against slabs from an optimization perspective. It can be modeled as a partitioning problem with multiple objectives and constraints. In this paper we formulate this problem as a cost-minimizing integer mathematical programming. A heuristic method based on tabu search is proposed. The method combines the swap, insertion and cyclic transfers neighborhoods to determine good approximate solutions. Computational results on production data from an iron and steel factory are presented and analyzed. Comparison with actual slab matching indicates that the proposed method could produce significantly better matching