If there is an emerging technology that has captured the attention of plant operators, it’s generative artificial intelligence (AI). As we head into the holiday season, generative AI has become either the best present anyone could ask for or the biggest lump of coal everyone should dread. When it comes to process control applications, one’s perspective probably depends on whether the data used to train a generative AI model is nice or naughty.
There are plenty of voices in the technology sector warning of the importance of high-quality data for productive AI decision-making. I recently attended a panel discussion where leaders such as Rick Kephart, VP of technology at Emerson’s Power and Water Solutions business, and Rita Wouhaybi, senior AI principal engineer at Intel, spoke extensively about the role of data in the future development of AI in industry.
The dirty truth about AI is that it can only be as good as the data it is trained on, they said. Large-language AI models, such as ChatGPT, have become popular over the past year, but to be successful in industrial operations, they must be trained with specific plant data, much like a digital twin.
In this issue, we highlight the FieldComm Group's 2023 Plant of the Year, and it’s an example of the necessary interplay of historical data and future, AI-augmented plant processes. Daikin Industries Ltd., in Osaka, Japan, recently completed a digital transformation of its Kashima flurochemical plant, and developed a system to capture historical data from its HART-enabled devices to train AI models to augment process control. The system not only helps current operators be more productive, but also helps ensure future generations of operators will have the same learning models available to them when they take over managing the plant.
Meanwhile, a quick Google search for “the importance of data in AI” will yield countless technical journal articles dating back two years or more that hammer home this same point: if you want AI to work for your operation, you better first learn how to be a data dynamo and not be a data dunce. Otherwise, there’s no telling what may take place on the plant floor.
AI models present a great opportunity for plant operators, but those systems must think as human operators would in any given situation. There’s only one way to ensure they do—ensure data quality. And, that data quality still originates in the same sensors, controls and historians that have been running process automation for a long time.
