This series of articles (Part I and Part II) intends to clarify on the relevant criteria to be considered when choosing either an optimization-based or a heuristic-based approach within the field of automated supply chain planning. This article is about the technical aspects, evaluating solution quality, flexibility and computational effort.
How do we get the perfect plan?
Before choosing a planning approach to determine the perfect production plan, you first need to define a metric for what “perfect” means for you individually. Is it the solution quality in terms of economic KPIs such as minimum costs or maximum revenues? Or rather technical figures like maximum throughput or a balanced resource OEE? Given your individual objective function definition, optimization will provide a mathematically proven optimal solution for any such cases. The downside of this favorable solution quality is the computational effort that comes with it. Depending on the complexity of a specific planning problem (e.g. measured as the number of products or production resources), the runtime of an optimization run can quickly become intractable (and memory requirements increase excessively).
In contrast to that, heuristics usually apply a set of business or planning rules/guidelines to approximate a sufficiently good solution without striving for an absolute optimum. The algorithmic design allows for the determination of a still feasible solution with far less computational effort. Given the dynamic production planning environments where e.g. rush orders or machine breakdowns may cause unexpected changes of the planning situation at any time, one may ask the valid question whether it is reasonable to undergo significant efforts to solve a snapshot planning situation to optimality. Would you rather accept a satisfactory planning result that can be determined quickly by heuristics? (For the differing key principles of heuristic-based and optimization-based planning, please refer to the first part of this series How to plan your supply chain Part I).
Complex equipment setup
Many production systems are characterized by a multi-stage material flow and/or a complex equipment setup with multiple alternative paths of products through the resource structure. Many potentially interdependent constraints constitute a huge challenge for the definition of a suitable set of planning rules and guidelines. In consequence, the design of holistic planning approaches based on heuristics practically becomes impossible and needs to fall back to problem decomposition. This may include the reduction of the number of products and resources to be considered, ignoring complex constraints or focusing on capacity bottleneck stages and will most probably involve manual replanning/rescheduling activities to ensure all constraints are met. In this context, optimization-based approaches provide one of their biggest benefits. Once you have represented all the characteristics of a production system in an optimization model, all constraints are considered simultaneously to derive a globally feasible solution across production stages (which ideally qualifies optimization as a one-click approach). Additionally to mastering complexity, you also get the flexibility to either easily extend the planning scope by adding products/resources or adjust constraints by introducing or adapting equations in the mathematical model. On the heuristic side, changes in the production system characteristics might entail the necessity to rethink the business/planning rules applied or the entire algorithmic procedure. As a follow-up remark on the second part of this series (How to plan your supply chain – Part II), please be aware that the definition of suitable objective function element weights or heuristic business rules is a key challenge to be resolved by the customer project team in order to manage the planning situation complexity and get the desired results.
AI is not the holy grail
Thinking about the above-mentioned complex planning environments, customers are often looking at the current trend of AI as an enabler for so-called one-button solutions. Just use AI and all your problems will be solved. No manual efforts anymore, just an intelligent little helper that always finds the best schedule. In real life that’s not the way it is. AI solutions can widely support in solving today’s planning problems such as identifying supply risks or updating planning parameters. However, the vast amount of solution alternatives in combination with the multifaceted interdependencies between planning constraints is just too complex to be solved when it comes to production scheduling. No data lake can support sufficient learning of such a solution. AI – at least at the moment – is just not the holy grail that everyone is hoping for.
This article completes the discussion of decision criteria for heuristic-based or optimization-based supply chain planning approaches from the user- and the solution quality-oriented perspective. As a key takeaway, we highlighted the trade-offs between the evaluated characteristics indicating the requirement for an individual assessment.
We thank Jens Rieder for his valuable contribution to this article.