Supply chain
Industrial AI solution development faces challenges
Chloe Liao, Taipei; Adam Hwang, DIGITIMES

Developing AI (artificial intelligence) solutions for industries face many challenges mainly because of lack of AI talent and availability of necessary data, according to experts.

And AI solutions are practically case-specific rather than for common use within a specific industry, the experts added.

Some market research organizations predict that AI will be mainly applied to four areas: financial services, medical/health care, retail operation and manufacturing. But manufacturing has the biggest potential for AI applications in Taiwan because of the country's industrial structure, followed by financial services and medical/health care, said Yu Shaw-shian, senior VP for government-sponsored Industrial Technology Research Institute (ITRI). Big data is the core to drive AI, and Taiwan's manufacturing industry consists of comprehensive supply chains that offer sufficient data for development of AI solutions.

Taiwna-based manufacturers fall into three categories in terms of readiness in collecting and accumulating operating data from production equipment, Yu said. The first is high-tech firms such as semiconductor and flat panel makers and/or large-size enterprises which have collected such data, Yu noted.

The second is makers which have collect insufficient data, but much of the data can be specially processed for use in developing AI solutions, Yu indicated.

The third refers to makers whose machines are of old models that cannot be web-connected to collect data, Yu said. Most of them are small- to medium-size firms and the third category accounts for over 95% of the total number of manufacturers in Taiwan, Yu noted.

External assistance, such as the use of smart machine boxes, is needed to enable such old machines to collect operating data, resulting in much time taken and less efficiency in developing AI solutions, Yu explained, adding thethird group pose the main difficulty in boosting industrial AI applications in Taiwan, Yu pointed out.

The time taken to accumulate enough data varies: for example, at least one year is needed for retail operation to reflect seasonal effects, and a few months to half a year for PCB makers, Yu said.

However, makers of different product lines may differ much in time taken to accumulate sufficient data and this is why it is difficult to develop AI solutions for common use in an industry.

While training in developing AI solutions is commonly based on deep learning, the purposes of using AI solutions vary from enterprise to enterprise, involving much higher multiplicity and customization than adoption of ERP (enterprise resource planning) or CRM (customer relationship management) systems, said CEO Chen Sheng-wei for Taiwan AI Academy.

© 2019 DIGITIMES Inc. All rights reserved.
Please do not republish, publicly broadcast or publicly transmit content from this website without written permission from DIGITIMES Inc. Please contact us if you have any questions.