The Synthesis of Autonomous and Adaptive Scheduling - Smart Agents for Smart Production Planning
Production planning is one of the most complex industrial challenges of the steel industry - it is roughly equivalent to the complexity of a Petri net with up to one million nodes! To address this complexity, metals manufacturers have developed several IT solutions and business processes. However, in addition to complexity, the production process is systematically affected by unpredictable events. In most plants today, responding to production disruptions is a human decision made by operators - often leading to response time issues that can cause production downtime and additional costs. The smart agent technology provides a remedy.
Steel production scheduling is a very complex business process that requires a complex collaborative solution. It involves different stakeholders, with common but also contradicting goals, such as:
- Keeping equipment busy and running
- Deliver finished goods in time, but don’t produce too early or too late
- Keep intermediate and finished stocks low but make sure you have enough
This makes production scheduling in the steel industry considerably one of the most difficult and complex industrial planning and scheduling problems. It involves several sequential production steps for each finished product, each step transforming the semi-finished material and needing different, often alternative, resources, transport, and warehouse systems (e.g. cranes, forklifts, piles). Furthermore, each step needs to fulfil critical, and sometimes opposing, production constraints.
Fighting the Complexity of Production Planning
To combat this complexity, metal producers have developed IT solutions and business processes. The efforts to meet the complexity are typically organized in a hierarchical decision workflow:
- Capacity planning
- Material allocation and combination planning
- Cross line campaign and production scheduling
- Detailed production line scheduling
- Transport planning
- Reactive schedule execution
This workflow is managed by many human operators such as capacity & material planners, line schedulers, production line operators, crane & transport drivers, warehouse managers, etc.
These human decision makers use software applications to help them make required decisions.
Besides the complexity of the production process itself, this production process is systematically affected by unpredictable events, such as equipment breakdowns, campaign restrictions, orders cancellations or changes, energy demand limitations, material unit unavailability, mechanical and chemical property misses. In most of today’s plants, reacting to and deciding on such disturbances is a human decision made by the operators; they have to detect the disturbance, analyze the possible impact and decide on corrective actions and communications as seen in our Stark Inc example aboce.
These human decision points, however, have some problems with reaction latency: Some operators are only available during office hours, people are not always focused and in general, humans are rather slow - the reaction time to disturbances could be minutes to hours.
The problem is that lagged or omitted disturbance handling can lead to production failures and extra cost. With smart agent technology, we can counter this.
Smart Agents Can Improve All Parts of the Reaction to Unplanned Events
Smart agents are small pieces of software focused on a very specific set of problems. They can automatically react to a wide variety of data sources and are capable of synthesizing a multitude of data inputs and assessing the impact of changes.
Smart agents are autonomous decision makers and can make independent, local decisions.
Like human beings, they can solve larger problems by communicating with other agents in so-called Multi-Agent Systems (MAS). We believe that smart agents can improve all parts of the reaction to unplanned events.
Imagine that smart agents could execute the following tasks:
- Tracking events related to each production order
- Observing the output quality of the casting of rolling process, triggering alarms if needed
- Observing the feasibility of a released caster or HSM schedule and triggering delays or rescheduling when needed 24/7
- Handling rescheduling and propagating the consequences to other smart agents
- Observing material movements to verify material availability requirements
In this world, everyday life at the Stark Inc. flat steel plant in Massachusetts would look different:
Explore the Nature of Your Supply Chain - Series
Part 4: The Synthesis of Autonomous and Adaptive Scheduling - Smart Agents for Smart Production Planning
Luc Van Nerom
Deputy Managing Director PSI Metals
After studying mathematics and computer science, he founded Artificial Intelligence Systems in 1986. The mission was to bring AI technology to the energy and process industry. Especially in the metals industry this crystalized into a number of production management optimizers today fully embedded into the PSI Metals portfolio. Nowadays at PSI, he focusses on innovation and product management. He is also managing director of PSI Metals Belgium.