Dec 31, 2020
About the Customer
The Engineering team, together with partners Pathmind and The AnyLogic Company, recently delivered an innovative Artificial Intelligence (AI) project to help a global Food & Beverage Manufacturing leader maximize factory output by making smarter decisions about production order sequencing, reducing average order processing time in a single plant by 16%.
The Customer’s Challenge
The factory being optimized had several production lines, each with different attributes and capacities. New orders came into the customer’s distribution hub every day. Those orders had different processing requirements, batch sizes and processing times. In addition, factory operators worked based on a set schedule, and they had different skill sets.
When handling and assigning production orders, a high number of moving parts had to be considered:
- Different production lines had to be fed with different orders every day;
- Orders were characterized by different code articles;
- The sequence of order execution had to respect certain constraints;
- Some order code articles needed to be assigned to a specific line;
- Every order code required a specific number of operators and time;
- Every production line had a minimum and a maximum number of operators that could be assigned.
The customer faced persistent bottlenecks and struggled to sequence orders in a way that would both saturate capacity and minimize costs, often running the risk of not fulfilling all orders within the required time window as a result. The factory needed a better solution that would increase productivity and reduce order processing time. The customer's initiative aimed to successfully apply advanced Manufacturing Simulation, Artificial Intelligence and Deep Reinforcement Learning methodologies in order to optimize the process of production order handling and management across the many lines, stations and operational teams within their main production facility.
About the Solution
The factory’s existing production order handling system was designed to process an order as soon as the necessary resources, i.e. the right production line and operators, were available. But that heuristic was unable to either reduce bottlenecks or to plan for them. The customer knew there was room for improvement. That's where Artificial Intelligence and Simulation Modeling came in to play.
Leveraging AnyLogic and Pathmind technologies, Engineering's simulation team set out to build a Digital Twin of the customer's existing manufacturing facility and processes, and then to train an intelligent model capable of automatically optimizing order sequencing, maximizing asset utilization and preventing bottlenecks. The policy was also trained to generalize, so that its logic could be applied to orders it had never seen before.
Our expert team used combinations of Artificial Intelligence and Machine Learning approaches, including Deep Reinforcement Learning, to train the learning agent and create the optimal policy through consecutive interactions with a white-box model of the system, encoding the stochastic behavior of production capacity and the uncertainty in demand. The Reinforcement Learning policy delivered and now running at the customer's site is able to intelligently control how production orders are matched with production lines in the factory. The policy resulted in a 16% reduction in order processing time for the customer compared to the factory’s best previous heuristic.
Artificial Intelligence for Industry 4.0
One of the main objectives of any manufacturing enterprise, or any other business for that matter, is to increase profits by maximizing output and minimizing operational costs. In this sense, Food & Beverage Manufacturing represents a highly strategic and competitive sector. To increase margins on specific products, effective management of factory processes, logistical operations, materials and production orders is needed. A smart digital approach speeds up time-to-market by optimizing asset utilization and guaranteeing quick reaction to changing needs and demands. In this light, enabling technologies such as Artificial Intelligence and Machine Learning are undoubtedly the next horizon for tomorrow’s industrial, manufacturing and supply chain leaders.
