In recent years, the working population is expected to decrease in developed countries, including Japan.
Difficulty of acquiring young human resources because of the declining birthrate, and the aging of engineers means that if these trends continue, the experience and know-how developed over many years will be lost due to the inability to pass on technology, and stable manufacturing process may become difficult to achieve.
Japan’s paper mills are no exception to this, and as it is difficult to acquire new human resources to work in the harsh conditions of high temperatures and humidity, and as experienced workers continue to retire, the reform of work style at production sites could become a major issue for future management.
As one of the solutions to these problems, we have developed “SmartPapyrus” to prevent defects and sheet breaks by visualizing machine dirt deposit using IoT, analyzing it using artificial intelligence, and using machine dirt deposit prevention technology.
In the process of developing this system, it was important to have information on what kind of defects occurred, “when”, “where” and “how many” during the operation in order to conduct a predictive analysis of defects and sheet breaks. However, the current defect detectors (WIS:Web Inspection System) are only used to determine whether defective products haven’t been shipped, and do not output information about where defects have occurred on the paper machine.
In order to identify the location of defect occurrence, experienced workers on site have to make decisions based on their experience using the defect information output from the WIS.
In order to make it possible for anyone to make quantitative decisions about the process, we developed an AI that incorporates the hunches and tips of experienced workers through image classification using deep learning, and developed a defect image classification system that supports operations based on the AI inference results.
This time, we will report our past efforts and future prospects.
View full abstract