Leveraging AI to Manufacture the Future

Professional services firm Accenture conducted a survey in 2018 on the adoption of AI targeting 500 industrial firms in six major economies. The results showed that, of the executives responding, almost 70% recognized the power of AI to influence their business’ futures, and displayed the determination to follow through. Only 16%, though, had actually taken the next step. A mere 2% were deploying large-scale AI-based solutions, while many companies that had taken action in some form either achieved results that were disappointing, or failed completely.

There are numerous reasons for these unsuccessful attempts. Whether it’s due to having adopted an “AI solutionism” approach—believing that simply introducing it will resolve issues—having miscalculated the financial and technological capabilities the endeavor would require, or lacking understanding of the nature of data collected or the ability to analyze or apply it, disappointing results stem from common causes: the inability to understand just what AI is, and a failure to identify issues that need resolution and adopt tools specifically suited to the purpose.

“For me, artificial intelligence is that activity devoted to making machines intelligent, and intelligence is that quality that enables an entity to function appropriately and with foresight in its environment.” ^*1
Nils J. Nilsson, Kumagai Professor of Engineering (Emeritus), Stanford University

Expectations are high for the potential of AI in the manufacturing industry. But energy and resource costs continue to climb, and new players are entering the market on the back of global technological advancement, particularly that related to statistics and pattern recognition, and big data platforms. Companies must continuously optimize cost performance and quality in order to remain competitive.

Before making any kind of financial investment, a company needs to establish its objectives, and involve personnel with extensive domain knowledge. AI is not a product; it is the science of enabling machines to develop the ability to imitate human cognitive processes, and to build and expand knowledge as more data is made available. But without people with in-depth knowledge of manufacturing processes, the data collected can never be analyzed or applied effectively.

AI combined with manufacturing industry domain knowledge can result in solutions that facilitate plant condition diagnosis and predictive maintenance, thereby protecting valuable assets and enhancing operational efficiency. Interpreting the data in a way that gives it value beyond itself requires considerable experience and domain knowledge. Companies that understand AI and skillfully craft and incorporate AI-based solutions will remain strong in an increasingly competitive global environment.

Yokogawa’s Komagane plant, which is responsible for the manufacturing of semiconductor parts called pressure sensors is located in Nagano Prefecture. There are hundreds of processes performed in the manufacturing of semiconductors. Microfabrication processing takes place, and as some of these processes are conducted internally, quality control is one of the key issues. For this reason, at the Komagane plant, numerous sensors were installed for each process to collect data, and the condition was judged while the data was being processed. As there were many processes that affected quality, however, analyzing the data from these processes and detecting the most complex, minute fluctuations was beyond the ability of human beings to perform.

Amidst all this, given the current popularity of AI, expectations within the company regarding the latest digital technologies were heightened. Engineers such as those in charge of products and manufacturing equipment maintenance, however, were not sufficiently familiar with the capabilities and possibilities of digital technology, and did not know what types of issues it could resolve. Likewise, R&D staff were unable to determine where their own technologies could best be put to use. So, the departments engaged in the exchange of considerable information and opinions regarding the special characteristics of a variety of AI tools.

At the same time, they set about identifying issues to be resolved using new technologies such as AI. Approaching it from the perspective of SQDC (safety, quality, delivery, cost), a performance-driven management protocol, they isolated key issues, to which they then assigned priority levels. They selected the problems with the highest likelihood of being solved using AI, and began to develop solutions based on the most suitable AI technologies available to them. Now, with a clarified team structure and clear objectives, the plant has successfully established a platform for co-innovation.

“By far the greatest danger of Artificial Intelligence is that people conclude too early that they understand it.” ^*2
Eliezer Yudkowsky, Co-founder and Research Fellow, Machine Intelligence Research Institute

At the Komagane plant, an AI-based analytical tool developed by Yokogawa is used to perform quality analysis on manufacturing process data. The number of attributes was calculated using signals from the sensors that had been installed for the purpose of ascertaining the conditions, and they have introduced a multi-dimensional analysis tool in order to reveal complex fluctuations that could impact quality. For the first time, through the application of massive volumes of IIoT data, a causal relationship was revealed between the attribute values and their electrical characteristics, enabling them to discover phenomena that would impact those characteristics. A new correlation was also found between production line data and quality. The AI-based analysis tool played a crucial role, as it assisted them in identifying the important items from large volumes of data, and in finding significance in data they had previously considered unimportant. The spread of IIoT will lead to continued growth in types of data and in volume—surpassing our capabilities to process—but AI will make the unveiling of multidimensional, complex causes relatively easy. AI characterizes changes that would go unrecognized by human beings even in graph form. Moreover, AI has begun to be applied on a broader scale, such as in the use of images in product and plant inspections, and in energy-saving plant operation by reading fluctuations in the outside temperature and operational forecasts.

AI is a tool, and it needs to be educated using the vast experience and knowledge acquired by humans. People with extensive domain knowledge and an understanding of analysis work together to teach AI to perform the required task, and provide the experience and insight to interpret the results to obtain maximum value. Humans must serve as kind of a broker in bringing AI to the world in a meaningful way.

Yokogawa, by accumulating case experience such as the results achieved at the plant in Nagano, and by fortifying its internal co-innovation structure, aims to develop a reinforcement learning algorithm to put into practical application at plants. The company will use its vast domain knowledge and industry experience. And, while co-innovating with its customers, it will apply AI strategically to the stabilization of quality and predictive maintenance for facilities, as well as to the improvement of processes and operations. This will lead to the achievement of swift, flexible management for its customers, to the optimization of their operations, and to the creation of new value.

---

References

*1 : Nils J. Nilsson, Kumagai Professor of Engineering (Emeritus), Stanford University, The Quest for Artificial Intelligence: A History of Ideas and Achievements, 2010
*2 : Eliezer Yudkowsky, Co-founder and Research Fellow, Machine Intelligence Research Institute, Singularity Hypotheses: A Scientific and Philosophical Assessment, 2012