Frontier Technologies in Industrial Operations 2025

Snapshot of sample case studies of virtual AI agents TABLE 2 Use case Challenge Solution Benefits Autonomous control agent for steel manufacturerSteel manufacturer KG Steel faced two main challenges: 1) high liquified natural gas (LNG) energy costs to operate furnaces and, 2) discrepancies in the product quality arising from a skill gap caused by an ageing workforce. To meet the desired quality of coils, operators must adjust the heating settings of furnaces while accounting for varying production environments. Inefficient heating process control causes excessive use of LNG. A South Korea-based AI software provider developed a deep learning model predictive control optimization model that executes “what-if” scenarios based on possible control patterns, and assesses them in a digital twin to select the optimal controls. Initially, the agent acted as adviser, providing operators with recommendations on optimal control settings. Partial automation of furnace operations was later achieved via system integration and direct feeding of agent output into the furnace control system.The agent has decreased LNG consumption by approximately 2%, while reducing differences in the product quality. Planning AI agent for global brewerA global brewer aimed to improve its planning process and forecasts. The current approach, known as post-game analysis, entailed continuously assessing past forecast inaccuracies and their root causes. This was tedious and required expert knowledge that dissipates over time.Post-game analysis is increasingly conducted by LLM composite agents that integrate a sequence of atomic agents to perform complicated exercises. They are trained on post-game “recipes” built by planning experts, learn continuously from feedback to improve results over time and can be used in cross-functional planning processes. Beyond productivity, the agents enhance knowledge and expertise in an organization.By using the agents of a US-based planning solution provider, the brewer achieved 70% touchless demand and supply planning. They additionally issued a resolution for feasibility checks. This percentage is expected to increase further as complex LLM/agent capabilities are added and further enhanced. Key success factors for adoption and scaling are the quality of the data, the ability to capture the decision- making logic of planners, plan explainability and trustworthiness of outcomes. Based on these criteria, the mid-term ambition is to achieve 90% automation scope and global scaling across markets. Autonomous quality control AI agentAs a globally recognized digital lighthouse,7 Siemens Electronics Work Amberg (EWA) has set the ambitious goal of achieving a first pass yield (FPY) exceeding 95% with a defect per million connections (DPMC) below 10 per production batch. This is a significant challenge, given that a circuit board can have up to 3,800 quality features to monitor. So far, the FPY target has been unachievable as employees have not been able to consistently make the right decisions given the time constraint and stress.To achieve this objective, EWA developed a patented autonomous quality control AI agent in collaboration with RIF Institute for Research and Transfer. This advanced agent, harnessing self- organizing maps, assists employees in correctly setting up the solder paste printer for the first production run. This reduces process times in a complex, multi-parameter task that typically requires significant experience. As process parameters are continuously improved, the agent adapts by accounting for parameter changes, resulting process behaviour and prior adjustments. This enables it to continuously learn and optimize the unknown behaviours of the solder printing process. After the testing phase, the agent will be able to autonomously adjust the parameter settings.Multiple studies have consistently demonstrated high-quality product output while simultaneously reducing the solder paste printer’s process time by up to 50% compared to the cycle time. The next phase of development involves transforming the agent into an edge application for rapid scalability. A key requirement for this advancement is integrating the quality inspection gates into the comprehensive digital twin of the process. Frontier Technologies in Industrial Operations 14