This paper proposes a deep neural network with module architecture for model reduction, and a cost function suitable for training the model. In the proposed model architecture, each layer is modularized for reducing the model by adjusting the number of layer. This feature allows the computational load of the model to be quickly switched. In order to maintain the accuracy of the reduced model even if it is not retrained, the cost function is defined as a weighted average of the errors of the model output over the number of layers. The effectiveness of the proposed method is validated through numerical examples for small tasks. Our implementation is available at https://github.com/sy-takano/modularized_dnn_for_model_reduction.

The surface quality of steel products is important in terms of both product appearance and material properties, requiring the installation of a surface inspection system for automatic inspection. However, it is extremely difficult to realize automatic inspection for steel products such as pipes, heavy plates and hot-rolled sheets which have an oxide film called “mill scale” on the surface, because the surface texture of the mill scale prevents detection of defects. Most of these products have been inspected visually by human operators. In order to solve this problem, the authors developed an optical surface inspection system utilizing a new technique called the “twin illumination and subtraction technique” and realized automatic inspection of rolled pipes in the hot condition by high detectability of surface convex-concave defects. This technique is able to detect only convex-concave defects and remove noise signals from the flat surface texture by illumination from two directions and subtraction. We discuss the new surface inspection system for heavy plate products by improving the proposed method. Specifically, the optimum optical system for high detectability was designed based on laboratory experiments and simulation of the reflection model, which clarified the optimal optical settings such as the incident angle and field of view. By using this knowledge in the design of the inspection system, we realized practical automatic surface inspection for heavy steel plates.

With the increased interest of manufacturing education and creative education, the number of education programs for experiential learning on mechatronics is on the rise. Most programs introduce the experiment and practical training with actual robots into education for the purpose of enhancing educational effects. However, the burden of preparing and maintaining the robot is not small and sharing an actual robot with some students increases the risk of infection from Coronavirus (COVID-19) in recent years. In this research we have developed a new simulator that supports education of mechatronics using AR technology. By using the simulation, the cost of preparing and managing the actual robot can be significantly reduced and the risk of infection disease is also reduced. In addition the AR technology makes it possible for students to make experience as they control the actual robots.