Bench Talk

AI agents coordinate energy decisions while meeting production and equipment requirements

Jiraporn/stock.adobe.com; generated with AI

Energy use in industrial equipment is interconnected. For example, reducing consumption in one part of a plant can affect operating conditions or production elsewhere. However, simply seeing a lower reading on an energy-consumption meter does not indicate whether the reduction is workable.

To assess a proposed change, engineers need to know current machine states, production schedules, and equipment operating limits. This information is stored in different systems and often changes during a shift. Artificial intelligence (AI) agents help bring this context together and assess the available options.

In the second part of our blog series on agentic AI’s role in industrial automation, we discussed the ways AI enables predictive maintenance in industrial equipment. In this final blog of the series, we define energy optimization, examine how AI agents could improve energy use and efficiency, and explore how an implementation could work in an industrial automation setting.

Understanding Industrial Energy Optimization

Industrial energy optimization is the practice of choosing how and when equipment operates to improve a defined energy outcome. A facility that aims to reduce electricity consumption, energy cost, peak demand, or emissions associated with its energy use still needs to meet production deadlines, maintain product quality, and stay within equipment and safety limits.

The target measurement will depend on the exact goal. For example, measuring kilowatt-hours per acceptable unit produced shows energy use relative to the output. Another key measurement involves identifying energy cost per production batch, which captures the effect of electricity rates on a particular cycle. Moreover, peak demand measures the highest power draw over a specific interval and idle consumption shows how much energy equipment uses when it is not producing any output.

These measures are not interchangeable. Moving a batch production to a lower-priced period reduces its cost without reducing the energy it consumes. But by prioritizing a lower peak in demand, total consumption remains unchanged. So comparing energy per unit also requires comparable production conditions; otherwise, a change in output could be mistaken for an efficiency gain.

Energy-monitoring systems record these values and alert human operators about the changes. But optimization goes further, using measurements, production schedules, and equipment constraints to compare operating methods for the plant. For instance, high idle consumption points to an opportunity, but shutting down the equipment helps only if the next operation can still start on time.

Using Agentic AI to Support Energy-Efficient Operations

An AI agent brings together energy-monitoring data from separate systems. The agent can call a forecasting model to estimate upcoming demand and an optimization model to compare possible operating plans. The forecast predicts what may happen, while the optimizer evaluates options, such as which approved equipment combination should run. The agent coordinates among different tools and incorporates the results into a recommendation.

The ability to revise a plan becomes important when the operating conditions change frequently. For example, a fixed energy-saving rule follows its predefined response, but a delayed batch makes the response unsuitable. The agent has to retrieve the new schedule, request a forecast, compare the available options, and update its recommendation. Plant personnel would still need to approve consequential changes, showing the necessity for a human-in-the-loop workflow.

Implementing Agentic AI for Industrial Energy Management

As an example implementation, consider an industrial plant with several production lines and shared compressors and chillers. The plant must meet its production target and keep the facility’s demand below a defined limit.

Smart meters and equipment controllers send time-stamped energy measurements and machine states to an industrial data platform. The production system gives the current schedule and output figures.

The AI agent receives the production target and energy objective. It gathers the current operating data and calls forecasting software to estimate the facility’s load over the next several hours.

If the forecasting model indicates a potential demand peak, the AI agent calls an optimization model to identify possible changes. These include changing a process within its completion window, reducing an idle period, or staggering the startup of high-load equipment.

The agent checks for proposed changes to the production schedule and equipment constraints. If the production order changes, the agent must update the analysis or refer the proposal to a human operator.

The implementation should begin in advisory mode, where the agent provides the expected energy effect, the affected equipment, and the constraints. The operator reviews and approves any specific changes. Then, confirmed schedules are applied through the existing manufacturing execution system (MES).

After the recommendations and verification process has proven reliable, the plant operator should consider allowing the agent to automatically submit narrowly defined, reversible actions within approved limits.

Balancing Energy Goals with Operational Constraints

Agentic AI is not designed to magically remove constraints on industrial energy decisions. Many aspects, such as production targets and equipment limits, will continue to determine possible changes. An agent helps integrate current plant data with forecasting and optimization models to identify feasible options.

The significance of using agents lies in coordinating the response, which includes preparing a recommendation, routing approved changes, and verifying that energy performance has improved without disrupting production. However, throughout the process, operators should retain authority over plant operation.