PC-SEP23-PG26.1_Layout 1 04/09/2023 15:56 Page 26
INDUSTRY 4.0/IIoT
AI ASSISTANTS IMPROVE ANALYTICS ACTIVITIES
By integrating AI into advanced analytics workflows, users can speed up deployment, conduct more complex analyses, and increase process optimisation insights, says Katie Pintar, senior analytics engineer at Seeq Corporation
I
n the era of Industry 4.0, process manufacturing organisations have invested in data-driven decision-making, embracing the implementation of advanced analytics solutions enterprise-wide to gain insights into manufacturing and industrial processes. With the launch of ChatGPT and, recently, GPT-4 to mainstream society, integrating AI to enhance many tasks has never been more popular or accessible. Applied thoughtfully, AI can amplify the impact of advanced analytics solutions by enabling complex workflows and accelerating time to value. ChatGPT/GPT-4 is a generative pre-trained transformer (GPT) model, a type of large language model (LLM). These models have exhibited a large leap forward for natural language tasks, and with the release of GPT-4, they are also useful in AI-assisted workflows. Using ChatGPT or another leading AI platform as an assistant in the space of time series data analytics can amplify subject matter expert (SME) workflows to increase productivity and time to insight. To be most effective, both SMEs and AI tools require high- quality data, contextual information, and a platform with the flexibility to implement and operationalise insights.
Advanced analytics applications are essential to this task because they centralise data originating from disparate sources, simplifying access, creating context, and implementing live updates. Context is particularly critical for AI tools’ effectiveness in analytics workflows.
Providing context to AI models like ChatGPT has become an important topic because many LLMs are not trained in real-time, potentially resulting in missed new information. For example, ChatGPT-4 is only trained on information on the internet up to September 2021. To provide context to models, relevant data can be included with a question—one type of embedding technique—when prompting.
Further improvements can be obtained by prompt engineering, which includes breaking down an objective into specific tasks or even providing certain phrases like “step by step” to boost accuracy. As advanced analytics
26 SEPTEMBER 2023 | PROCESS & CONTROL
applications integrate AI into their toolsets, SMEs must be able to contextualise the data before invoking Al.
To implement advanced analytics workflows with AI input, applications need the flexibility to complete complex analyses and operationalise the results (Figure 1). Without flexibility, the integration of AI into advanced analytics workflows is limited, and without a way to seamlessly share and communicate the results of an analysis, the impact is minor. Before these workflows can be truly integrated, companies must overcome many obstacles including costs, trust, and data privacy. Due to high compute demands, training and running models for inference can be expensive. Ineptly-tuned models can be inaccurate, or even make up information. And companies need to understand how and where data is being saved for privacy purposes. This is why understanding and selecting appropriate commercial vendors or open- source LLMs to integrate into system architectures is critical. The following use cases highlight practical instances where SMEs drove analysis and implementation using AI assistance.
Asset monitoring at one process manufacturing site is enabled using an advanced analytics solution to define, analyse, and organise alarm conditions, allowing operations to focus on the most critical issues. Integrating ChatGPT into the workflow helped sift through numerous user comments, incident reports, and work orders to summarise patterns and rapidly determine process optimisations (Figure 2).
Leveraging Seeq, an advanced analytics solution, a user built a fully embedded tool to perform partial least squares (PLS) modeling by providing a single request to the ChatGPT model. General knowledge of the task was necessary to optimise the ChatGPT prompt and verify the code, but most coding was performed automatically by the ChatGPT model, enabling creation of the tool. This accelerated the user’s time to value, providing a nearly seamless experience with complex workflows. Additional workflows include embedding a coding co-pilot into the Python
Figure 1 (top): Seeq operationalises the results of advanced analytics to drive decision-making in this regional manufacturing dashboard. Figure 2: A ChatGPT assistant in a Seeq alarm and event monitoring application is used to summarise patterns to help SMEs optimise decision-making. All figures courtesy of Seeq
application, enabling the SME to access ChatGPT capabilities within the application. A process engineer was monitoring a controller that provides an error calculation, but they were not sure how to use percent error to create relevant key performance indicators (KPIs) to monitor and predict failures. Using ChatGPT and proper prompting techniques to provide context, they determined several appropriate KPIs and calculations, ultimately selecting “mean absolute error” for the analysis. The engineer now calculates mean absolute error using a flexible advanced analytics solution and uses it to monitor controllers in near-real time to predict failure and monitor performance. AI is a game-changer for process manufacturing companies wanting to amplify their analytics to unlock quicker and more impactful process optimisation. Stacking AI tools atop advanced analytics applications can speed up deployment and bridge the gap between users’ ideas and the required development work, quickly bringing analytics activities and process insights to life.
Seeq
www.seeq.com
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45 |
Page 46 |
Page 47 |
Page 48 |
Page 49 |
Page 50
