Modeling the digital factory

But new ideas are already ripening in the development labs of software developers. Orsoft, one of the leading development and consulting companies in the field of planning software, provides an insight. Viewing planning as a holistic tool for controlling company processes is at the top of the agenda. Autonomous processes are accordingly included in the entire store floor level and modeled on the basis of digital twins.

Before delving deeper into the subject, some terminology needs to be clarified. In classifying the term artificial intelligence (AI), Orsoft uses the definition provided by supply chain management market research specialist, Gartner. AI applies advanced analytics and logic-based techniques, including machine learning, to interpret events, support and automate decisions, and take action.

Machine learning (ML) is a form of AI that aims to teach an algorithm through repetitive training in such a way that it can perform tasks independently. Unlike conventional algorithms, the recognition of (data) structures is not predetermined by an implicit model structure, but is left to the "machine" autonomously.

Autonomization of planning processes

With the patterns recognized from ML, the forecasting quality of time series-based processes attains a completely new quality. The highest goal of ML to be strived for so far is autonomization. In differentiation to automation, Fraunhofer IESE describes the terms as follows: If it is known how a system behaves in a certain situation, it is automation. If this is unknown, we speak of autonomization.

A smart extension of existing Enterprise Resource Planning (ERP) systems by Advanced Planning Systems (APS) is aspired. Since ERP is not explicitly designed for planning tasks - for example, only rudimentary inventory controls or simple planning processes such as MRP runs can be mapped - additional advanced planning systems are often used to represent and optimize transactional business processes. However, these no longer meet the expectations of today's planning reality.

The systems, which are often not very agile, do not map entire supply chains or the possibility of ad hoc plan adjustments close to the batch size of one. With its software tools for strategic-tactical and operational planning, Orsoft has always pursued the integrative end-to-end approach to planning - that is, all planning instances are modeled holistically, intelligently linked with each other, and diverse simulation and forecasting options are created. These characteristics are the basic prerequisite for the application of (partially) autonomous planning instances, the third major disruption in supply chain management after ERP and advanced planning systems.

Digital twins

Autonomous planning instances are to be modeled with the help of a digital twin. What is a digital twin? Analogous to genetics - where identical twins share identical DNA - the digital twin shares specific functional properties with its real brother. Transferred to industrial reality, this could be, for example, the store floor and its digital copy in the area of production or a digital supply chain twin in the area of planning.

The digital supply chain twin is the basis for the autonomization of planning. With it, processes become more networked, efficient, agile, faster and more transparent. The more complex the value chains - a plant with a sales unit or a comprehensive network of production, purchasing and sales processes makes a big difference - the more necessary is the real-time synchronization of data between the planning instances and the central ERP. 

A uniform data layer with maximum detail accuracy, data integrity and response speed can be provided via the modularly expandable and interface-open PaaS platform Edge.One. This allows numerous processes from the store floor and pot floor to be mapped and integrated into the central platform approach in the sense of the digital factory.

Advanced Analytics

Advanced analytics is no longer limited to past descriptive analysis of data (the "what happened?" question), but focuses on predictive ("what will happen?") and prescriptive ("how can I make it happen?") analysis.

With advanced analytics, scenarios can be mapped automatically and executed based on predefined goals or goal hierarchies - such as maximum customer service, customer segmentation, margin-optimized production program, etc. Prioritization rules help to resolve conflict situations in the best possible way and automate them using machine learning. Depending on the complexity of the plan optimization, the human factor can be partially or completely deprived of decision-making authority. 

Based on ML analysis of time series, reliable forecast values are provided for demand planning. In this way, fluctuations in the supply chain can be reduced and subsequent processes improved. The conclusions for inventory management are equally evident: The reduction of necessary safety stocks to cover demand fluctuations is another important goal. With the help of AI, inventory planning can be switched to dynamic safety stock. Reactive scheduling is a buzzword in the field of AI-supported planning. Here, the planning run is described as an agile process that is open until the last minute. The goal is to be able to react to changing conditions - ad hoc orders from an A customer, reorders, delivery bottlenecks, employee/plant outages, or the like. Continuous learning from decisions made in the past continuously "feeds" the AI with data. An always up-to-date planning engine in the background, fed by the live system, helps to ensure that the initially calculated plan only has to be adjusted marginally and continuously. 

Machine learning methods also provide interesting features in the area of conflict management. Based on the user's own previous decisions and the analysis of the results, suggestions for solutions are created and, if desired, autonomized. With the help of the comparison of planned production and real feedback from the store floor, the quality of the plan can be evaluated and fed back into the modeling of the planning algorithm - also with regard to the implementation of maintenance forecasts. Alarm functionalities are also (partially) automated and self-learning through AI-supported decision processes.

SAP environment

Autonomization of Advanced Planning and Scheduling (APS) in the SAP environment is an Orsoft use case. The customer's task was clearly defined: Touchless autonomization of operational planning with the goal of implementing a system that works autonomously in the normal case - i.e. in the so-called happy flow - and that can automatically involve planners when certain events occur. The following additional constraints were agreed upon: automatic (iterative) interaction with SAP, strong configurability of the data model, automatic reaction to specific, freely configurable events, and the execution of functions on a separate application kernel, which allows the software to be operated even without logged-in users.

In order to do justice to an autonomization of the operative planning, goals of the planning run, which are represented as given prioritization rules - also graded as a sequence of priorities - must be defined. Like the boundary conditions, these can be set in detail. All SAP fields - including those of referenced objects, for example the customers for the sales order of a planned order - can be configured both manually and automatically and ranked by maintaining limit and exact premise values. In this specific case, the consideration of material availability/reprocurement times, personnel presence and qualification were specified by the customer as explicit boundary conditions. 

Thanks to the bidirectional real-time synchronization with the master and transaction data in SAP, the plans created are always aligned in terms of capacity at the plant and material level. Orsoft applies a complex MRCP (Material Resource Capacity Planning) run with finite, i.e. limited available, resources without overloads, which is not available in SAP. In the actual use case, material provisions from customers are also integrated into the planning run. A CTP (Capable to Promise) immediate check implemented by Orsoft and interacting with SAP acts as a trigger for planning autonomization. Only when there are target conflicts or missing resources on material and capacity level, automated alarms are displayed and sent as e-mail alerts.