"Fundamental technology" is the foundation of Fujikura's technology that supports "Technology platform". We are researching and developing common technologies that can be applied to all technology platforms. In recent years, we have been working on the application of AI technology, which has made remarkable progress, to manufacturing.
Fujikura started research on deep learning in fiscal 2015 with the aim of applying AI to manufacturing. Deep learning is an AI technology that uses multiple layers of artificial neurons that are mathematically modeled after human nerve cells. By applying it to image judgment, it is possible to judge images with high precision equal to or higher than that of humans.Although there are few companies where AI is applied in the manufacturing field, Fujikura started applying it to manufacturing lines in fiscal 2018, and has been gradually expanding it according to the roadmap shown in Fig. 2.
Figure 1. Image recognition with multilayer neural networks
The image to be judged is input to the input layer in pixel units, and the characteristics of the image are extracted while the multidimensional information is reduced to a small dimension in the middle layer. Then, based on the characteristics, the reliability of each judgment type is output to the output layer.
Figure 2. Roadmap for deep learning efforts
The horizontal axis shows four kinds of information in order of effort: image, numerical value, control, and language, and the vertical axis shows deep learning technology.
An example of application to manufacturing is wafer appearance inspection of laser diodes, which are the one of the main components of fiber lasers. In the past, visual inspections were carried out by highly skilled engineers, but now, unmanned systems using deep learning have achieved stable quality and reduced manufacturing costs.We are also performing of the whiteboxing of AI when applying it to manufacturing. In addition to the design of an AI model that maintains high accuracy, the inference grounds of AI can be visualized (Figure 3), and it is possible to confirm where AI paid attention. In this way, we are developing an AI that can explain from the user's point of view.Going forward, we will not only improve our manufacturing processes, but also contribute to the provision of more value-added products and services.
Figure 3. Visualization of the basis of reasoning in a laser diode chip
The point of deep learning paid attention for classifying judgement is shown in warmer tones.
We strongly support the Fujikura Group's manufacturing activities by using our material technology that we have cultivated over a long period of time. Efficient development of material technology is essential to contribute to the construction of a sustainable society by realizing products that take safety and the environment into consideration. We try to improve the efficiency of material technology by introducing materials informatics, which applies information science to materials science. We continue to support the Fujikura Group's products and manufacturing through our core material technology.
In manufacturing, accurate object understanding, careful & accurate object observation & measurement, and reasonable object interpretation are keys to early problem solving and the creation of new value. In Fujikura we provides powerful support for research and development and product manufacturing by making full use of "observation" and "measurement" such as advanced observation technology covering the size range from nano- to macro-meter, highly accurate and reliable analytical evaluation technology developed over many years. We also develop new analysis and evaluation technologies to contribute on research and development and business promotion.
It is common knowledge CAE (Computer Aided Engineering) is effectively used in product development and failure analysis. CAE by computer simulation is also very useful in the development of manufacturing technology and processing technology.Computer simulation of optical fiber preform manufacturing requires analysis by "Multiphysics" that simultaneously calculates fluid analysis with burning and chemical reaction analysis of raw material SiCl4. Since “Multiphysics” analysis requires complicated calculations, improving the accuracy of analysis is a major technical issue.At Fujikura, we feed back actual manufacturing data to analysis, accumulate unique CAE technology know-how to improve analysis accuracy, and apply CAE to process development of optical fiber manufacturing.
Our company has manufacturing bases in countries all over the world, as well as various product groups. However, if we depend on human work, we cannot ensure the quality of highly accurate and fine products such as electronic components. Therefore, even in manufacturing in countries where labor costs are low, automatic assembly equipment and inspection equipment are essential to realize high quality manufacturing suitable for each product. Automatic assembly equipment requires high-speed, high-precision positioning, transportation, and development of various methods. Inspection equipment also requires image processing technology and AI technology. These advanced equipment are produced in house, with automatic collection and utilization of various manufacturing data, and bring us higher productivities.
In the field of manufacturing high precision products, Precision Parts, Fixtures and Dies are used to minimize process variations, achieve high productivity and high yield. Our company group makes full use of high-precision cutting, grinding, and EDM technologies to manufacture various types of dies, including punching, molding, and progressive feeding, thereby improving product accuracy. This cutting technology is applied to the manufacture of products, and we manufacture tens of millions of electronic components annually with high process capability by own design cutting-tooling. In the R & D field, we are developing machining methods for difficult-to-cut materials such as heat-resistant alloys and rare metals, as well as new materials, by making full use of various high-precision machines, including high-speed machines capable of machining with sub-micron precision. In the future, our company will continue to develop next-generation laser processing technology using a high-precision laser processing machine designed and manufactured in-house that utilizes the company's strength in high-power laser transmitters.