Following all the hype surrounding ChatGPT in late 2022, the possibilities offered by AI have become much more apparent, including for companies. It's all made possible by the application of generative AI and, in particular, the development of large language models (LLMs) based on virtually limitless quantities of training data on the freely accessible internet.
However, experts agree that generative AI is only a smaller subset of AI, and that in some specific areas, training data is not as widely available in the form of texts published on the internet. In such areas, monitored machine learning (ML) continues to be an important approach alongside LLMs.
The applications of AI are plentiful
"On this basis," says Bigvand, "there are a number of applications in production plant engineering that can benefit significantly from the use of AI." These include, for example:
- automatic generation of data models and diagrams using LLMs, i.e. automatic generation of a component's data model
- automatic correction of data models and diagrams, i.e. suggested corrections for the data model based on similar tokens identified by the LLM
- the HAZOP (Hazard and Operability) risk assessment, i.e. identification of constellations that can be considered hazardous
- migration of old documents using image processing and ML models.
AI can do a lot, but human experts are also needed
Similarly, AI can learn to 'understand' diagrams. It can be trained to classify the components depicted in PDF or PNG files. AUCOTEC is taking advantage of this and working on a unique form of support for projects seeking to transfer existing documentation to the data-centric Engineering Base software. The aim is to migrate all diagram types from the areas of process, instrumentation & control, electrical and hydraulic engineering – and generate a data model with hotspots for navigation purposes alongside the PDF files. "However, experienced experts are still needed to carry out checks and improve the AI model," explains Bigvand. "Several rounds of corrections and fine-tuning are required for each dataset." The training is worth all the effort, as it will then be possible to create object models very efficiently from plant documentation that is decades old. Doing so will greatly facilitate maintenance and modernization work. "While much of this is already possible for machine-readable formats such as DWG in Engineering Base, AI will breathe new life into documents that have long been considered obsolete," says Bigvand.
"Structured" data on a large scale
"At present, when engineering tool suppliers talk about their tool's AI functionality, they are usually referring to an advanced search feature that can handle large amounts of text and provide meaningful answers, lists or components," continues Bigvand. "At AUCOTEC, however, we are certain that AI is capable of much more in the field of plant and mechanical engineering if 'structured' data is available on a large scale." And, as Bigvand concludes: "Being able to capitalize on the capabilities of AI will be crucial for staying competitive."
Engineering Base seems to fit the bill perfectly here: open for integration, data centric across different disciplines, and exceptionally consistent and transparent too.