Machine intelligence for the optimization of supply chain networks

Finalisation of Move

The Move project, launched in October 2020, was successfully completed at the end of last year. Fraunhofer IML collaborated with a consortium of research and industry partners from the OWL region (Ost-Westfalen-Lippe) to develop methods and tools for using artificial intelligence (AI) methods. These methods can help companies analyse and optimize their supply chain configurations, including scheduling parameters, delivery dates, and sales forecasts, in a sustainable manner. Generated AI task models, and process model were tested in pilot projects with industry partners. Additionally, generic solution models were created for broad transfer.

The challenges of increasing complexity in supply chains

The supply chain network of manufacturing companies is becoming more complex due to internal and external factors. As the number of partners increases, so do the external influences and uncertainties on the supply chain, which can lead to disruptive events.  Additionally, cost pressures and expectations for delivery dates continue to rise. Today's entrepreneurs face additional challenges due to current trends and a high degree of individualisation, coupled with a decreasing product life cycle. To overcome these challenges, it is essential to have a reliable forecast of customer requirements and to monitor fluctuations in demand along the supply chain.  It is important to continuously evaluate the parameters set in planning within the company's IT systems and adapt them to changing environmental influences. The digital transformation of process-oriented supply chains is leading to ever-increasing volumes of data. This potential should be effectively utilised to meet the challenges in supply chain management.

Utilisation of enormous amounts of data

The use of AI processes has the potential to solve these problems through automatic analysis, evaluation, and optimization. The MOVE research project developed and tested practical methods and tools over three and a half years in close collaboration with industry partners. To use AI processes and methods effectively, the first step is to understand and utilize the enormous amounts of data. The companies were able to integrate AI processes into their supply chain networks thanks to the research project. The project focused on generating and integrating expert knowledge into simulative, data-driven, and hybrid tools.

Fields of action of the problem

The Move research project considered three fields of action for optimising supply chain networks through the successful use of artificial intelligence.

Specification of interdependencies: To develop practical and beneficial models for optimizing supply chain networks, it is essential to understand and specify the logistical interdependencies in supply chain management. Also, to create models, effective communication between domain and AI experts is required.

Specification and expansion of the IT infrastructure: A high-performance IT infrastructure is essential for efficient utilisation of the existing database. This forms the link between the physical and virtual supply chain.

Selection and development of AI methods: Supply chain networks are optimised and analysed using models and methods from the fields of simulation, AI and operations research.

optimization of processes in logistics with AI and digitization
© Fraunhofer IML

Realisation in three pilot projects

The methodology's application experience is being processed into solution models and a process model for transfer through three pilot projects. To achieve this, the CRISP-DM model was expanded to include aspects of consensus building of expert knowledge and communication with AI specialists in the form of a specification technique. Another important focus was the detailed inventory of existing data for developing customised AI and simulation solutions. Assistance modules were developed for three pilot projects to overcome challenges in demand and delivery date forecasting, as well as parts tourism inaccuracies.

Parts tourism: As part of the project, Fraunhofer IML focused on recording and reducing the tourism of an industrial partner. Recommendations and several simulation models using the internal simulation tool OTD NETWORK were created to analyse different network alternatives. By linking the simulation models with an algorithm configurator, they proposed and implemented a cost-optimized parameterization of selected inventory parameters. By comparing the optimized parameterization for reducing partial tourism and its impact on the overall network efficiency with the manually parameterized alternatives considered during the planning process, the potential benefits of these two linked technologies can be demonstrated.

Inaccurate demand forecast: As part of the project, the original sales forecast of an industrial partner was analysed. Also, the customer's forecasts and external factors were considered, such as industry-specific or economic developments, using tried-and-tested AI algorithms to improve the forecast values.

Inaccurate delivery date forecast: To ensure timely deliveries and maintain customer satisfaction, accurate delivery date forecasting is necessary. Two different approaches were used, both involving the analysis of historical data. Firstly, conventional statistical and stochastic methods were combined in an expert system to develop forecasts. Secondly, AI models were used to analyse and evaluate sub-process steps.

Role of Fraunhofer IML:

The Fraunhofer IML's Supply Chain Engineering department has significant expertise in practical research on supply chain management. We regularly use service-oriented IT solutions for contract research, which deliver promising results in strategic and tactical planning and operational control. These solutions are based on technological instruments such as the OTD NETWORK simulation tool, developed within the department for discrete-event simulation of supply chains. They are particularly useful in availability management and demand and capacity management. Based on experience and requirements from both the use cases and the method concept, we have further developed the OTD NETWORK simulation tool.