Advancements in technology have made deep learning a hot research area, and we see its applications in various fields. Its widespread use in silicon wafer defect recognition is replacing traditional machine learning and image processing methods of defect monitoring. This article presents a review of the deep learning methods employed for wafer map defect recognition. A systematic literature review (SLR) has been conducted to determine how the semiconductor industry is leveraged by deep learning research advancements for wafer defects recognition and analysis. Forty-four articles from well-known databases have been selected for this review. The articles’ detailed study identified the prominent deep learning algorithms and network architectures for wafer map defect classification, clustering, feature extraction, and data synthesis. The identified learning algorithms are grouped as supervised learning, unsupervised learning, and hybrid learning. The network architectures include different forms of Convolutional Neural Network (CNN), Generative Adversarial Network (GAN), and Auto-encoder (AE). Various issues of multi-class and multi-label defects have been addressed, solving data unavailability, class imbalance, instance labeling, and unknown defects. For future directions, it is recommended to invest more efforts in the accuracy of the data generation procedures and the defect pattern recognition frameworks for defect monitoring in real industrial environments.
In this paper, we present a novel deep reinforcement learning (DRL) based method that is used to perform multi-robot task allocation (MRTA) and navigation in an end-to-end fashion. The policy operates in a decentralized manner mapping raw sensor measurements to the robot’s steering commands without the need to construct a map of the environment. We also present a new metric called the Task Allocation Index (), which measures the performance of a method that performs MRTA and navigation from end-to-end in performing MRTA. The policy was trained on a simulated gazebo environment. The centralized learning and decentralized execution paradigm was used for training the policy. The policy was evaluated quantitatively and visually. The simulation results showed the effectiveness of the proposed method deployed on multiple Turtlebot3 robots.
1 January 2020 in Journal of Physics: Conference Series
Developing algorithms for multi robot systems to reach target positions and navigate safely in the environment is an open field of research. Most systems treat Multi Robot Task Allocation (MRTA) and Multi Robot Path Planning (MRPP) as two separate steps each with its own set of algorithms in which the MRTA algorithm assigns each robot to a task and the MRPP algorithm guides each robot through the environment towards the assigned goal position while avoiding both static and dynamic obstacles. In this paper, we present a method that combines both steps by using a deep reinforcement learning model. The model consists of a decentralized sensor level policy which outputs the robot’s velocity to guide it through the environment towards the selected goal position and avoiding collisions. The model was trained in a simulation environment and all the robots are homogenous differential drive robots. The objective is to ensure that each robot reaches a unique goal position with the number of goal positions is equal to the number of robots. The results of training the policy in an environment is presented with both static and dynamic obstacles with four robots and four goal positions.
27 September 2019 in Journal of Mechanical Engineering and Sciences
The vast impact on machinery that is rooted by bearing degradation thus pinpointing bearing fault diagnosis as indubitably very crucial. The research is innovated to diagnose the fault in bearing by implementing deep learning approach which is Convolutional Neural Network (CNN) that has superiority over image processing and pattern recognition. A novel model comprises of Gabor Transform and CNN is proposed whereby Gabor Transform is utilized in representing the raw vibration signals into its image representation. The CNN architecture is augmented for a better accuracy of the bearing fault diagnosis model. To date, the method combination has never been deployed in establishing fault diagnosis model. Plus, the usage of Gabor Transform in mechanical area especially in bearing fault diagnosis is meagrely reported. Scant researches in mechanical diagnosis are dedicated to work on the image representation of the vibration data whereas the CNN works better when fed by images input due to its unique strength of CNN in images processing and spatial awareness. At the end of the research, it is perceived that the proposed model comprises of Gabor Transform and CNN can diagnose the bearing faults with 100% accuracy and perform better than when CNN is fed with raw signals.
Halal requirement in food is important for millions of Muslims worldwide especially for meat and chicken products, insuring that slaughter houses adhere to this requirement is a challenging task to do manually. In this paper a method is proposed that uses a camera that takes images of slaughtered chicken on the conveyor in a slaughter house, the images are then analyzed by a deep neural network to classify if the image is of a halal slaughtered chicken or not. However, traditional deep learning models require large amounts of data to train on, which in this case these amounts of data were challenging to collect especially the images of non-halal slaughtered chicken, hence this paper shows how the use of one shot learning [1] and transfer learning [2] can reach high accuracy on the few amounts of data that were available. The architecture used is based on the Siamese neural networks architecture which ranks the similarity between two inputs [3] while using the Xception network [4] as the twin networks. We call it HalalNet. This work was done as part of SYCUT (syriah compliant slaughtering system) which is a monitoring system that monitors the halalness of the slaughtered chicken in a slaughter house. The data used to train and validate HalalNet was collected from the Azain slaughtering site (Semenyih, Selangor, Malaysia) containing images of both halal and non-halal slaughtered chicken.