• Written by Azhar Aliyar, Director

In theory, manufacturing systems should hum along smoothly, with each component working in harmony to produce high-quality products on schedule. In practice, however, the reality is often far from this ideal, with changeover losses, multi-plant comparability issues, and an ageing asset base all contributing to inefficiencies and downtime. This is where agentic AI workflows in manufacturing can make a real difference, by taking multi-step action to mitigate these issues and keep production running smoothly.

Manufacturing complexities

One of the main challenges in manufacturing is the need to balance production schedules with maintenance and changeovers. When a production line is shut down for maintenance, the loss of production time can be significant, and the impact on overall efficiency can be substantial. Furthermore, when comparing the performance of different plants, it can be difficult to account for variations in equipment, processes, and personnel. An ageing asset base can also lead to increased downtime and maintenance costs, as older equipment is more prone to failure and requires more frequent repairs.

A shift supervisor checks the maintenance log to plan the day's production, taking into account the scheduled downtime for maintenance and the potential for changeover losses. By using agentic AI workflows, the supervisor can get a better understanding of the production schedule and make more informed decisions about when to perform maintenance and how to minimize changeover losses.

Agent capabilities

So what separates an agent from a chatbot? The key difference lies in the agent's ability to use tools, maintain state, and work towards a goal. A chatbot is simply a program that responds to user input, whereas an agent can take action on its own, using the tools and systems available to it to achieve its objectives. In a manufacturing context, this means that an agent can use systems such as ERP and MES to optimize production schedules and minimize downtime.

For example, an agent can use the ERP system to check the production schedule and then use the MES system to adjust the production line configuration to minimize changeover losses. This level of automation and optimization can have a significant impact on overall efficiency and productivity.

Human approval

While agentic AI workflows can automate many tasks, there are still situations where human approval is necessary. Removing the human approval step prematurely can be a mistake, as it can lead to unintended consequences and errors. For instance, an agent may optimize a production schedule to minimize changeover losses, but in doing so, it may also create a situation where a critical maintenance task is delayed, potentially leading to equipment failure.

Most people get wrong the idea that automation and AI can replace human judgement entirely. They are wrong, as human approval is still necessary to ensure that the agent's actions align with the overall goals and objectives of the organization. By keeping a human approval step in the process, manufacturers can ensure that the agent's actions are aligned with their goals and objectives, and that any potential issues are caught and addressed before they become major problems.

Failure modes

An agent fails differently from a model, as an agent's failure can have a direct impact on the physical world. When an agent fails, it can cause equipment to malfunction, production to be delayed, or even safety issues to arise. It is therefore critical to log and analyze agent failures, to understand what went wrong and how to prevent it from happening again in the future.

By logging agent failures, manufacturers can identify patterns and trends, and make adjustments to the agent's programming and configuration to prevent similar failures from occurring. This can involve adjusting the agent's goals and objectives, or modifying the tools and systems it uses to achieve its objectives. For more information on how to implement agentic AI workflows in manufacturing, visit our website.

Limitations

Agentic AI workflows are not a silver bullet, and there are situations where they are not the best choice. For example, in situations where the production process is highly variable or unpredictable, an agent may struggle to optimize the production schedule effectively. In these situations, other approaches, such as machine learning or traditional automation, may be more effective.

If you are considering implementing agentic AI workflows in your manufacturing operation, but are unsure if it is the right choice, contact us to discuss your options and determine the best approach for your specific needs. By understanding the limitations and potential applications of agentic AI workflows, manufacturers can make informed decisions about how to use this technology to improve their operations and stay competitive in a rapidly changing market.