Building supply chain resilience through AI, dynamic SLAs and decentralised learning
In M4ESTRO, intensive research focuses on the use of AI techniques (e.g., Machine Learning) to identify factors that trigger supply chain disruptions from external data sources such as the operation of global supply chains and social media.
Our M4ESTRO colleagues at Netcompany-Intrasoft (INTRA) take advantage of various open source data feeds such as YouTube API and Yahoo Finance to develop data analytics services that identify supply chain disruption indicators (e.g., Financial Market Turbulence Analysis Service, Natural Hazards Monitoring Service). They also consider additional supply chain disruption indicator categories such as increased raw material prices and earthquakes. The analytics services that have been designed and developed are broken down to smaller components that are called processors (e.g., Anomaly Detection in Raw Material Prices, Keyword Matching). The latter are individually designed to provide the service’s detection functionality. Furthermore, a set of demonstration scenarios have been identified - including Covid-19 lockdowns, earthquake in Turkey in 2023, and PG&E bankruptcy- to demonstrate the functionality of various data analytics services and of their data processing components.
Service Level Agreements (SLAs)
INTRA has also a leading role in the development of M4ESTRO services that boost manufacturing resiliency at the equipment, data services and data exchange levels. Specifically, with the target being the greater resilience of manufacturing enterprises in the event of large-scale disruptions, M4ESTRO is working towards creating a federated data management environment where supply chain stakeholders can share data in a trusted way. Moreover, the project is specifying adaptive control strategies for manufacturing resilience, which will be implemented over the federated data management infrastructure. In this context, INTRA is leading the specification and implementation of the necessary Service Level Agreements (SLA) between the supply chain participants, which drive the implementation of the resilience strategies at the manufacturing value chain level.
During the last couple of months, INTRA has selected the smart contract technology to be used for managing dynamic SLAs. Moreover, they have designed a data model for the SLAs and provided detailed specifications of the elements and artifacts that comprise an SLA in the M4ESTRO context. Also, we have designed the SLA exposed services and the SLA monitoring agent, which are essential elements of the SLA management functionalities. Our next steps include the deployment of a proof-of-concept blockchain infrastructure that will be used to support the implementation of SLAs in a decentralized environment. This infrastructure will be tested in a number of already specified scenarios for dynamic SLAs, which will be modelled and implemented as smart contracts. We also plan to align the SLA concept with the concepts of standards-based data spaces i.e., to consider an environment where manufacturing resilience related data are exchanged subject to the established SLAs.
Federated learning
INTRA is also designing and implementing a federated learning infrastructure, which is destined to endow the M4ESTRO federated environment with extra intelligence. Federated learning will enable supply chain participants to learn resilience strategies based on available data and past/historic information. During the past weeks, they have emphasized on modelling the M4ESTRO manufacturing resilience scenarios as federated learning problems. To this end, they have determined the inputs and outputs of the federated learning nodes. They have also finalized an initial version of the federated learning framework (see Figure 1 for its sequence diagram), which utilizes REST API connections. This is to be integrated with the project’s Data Spaces for trusted data sharing across stakeholders.
Bottom Line
Overall, these activities are actively contributing towards the development of a robust, adaptive, dynamic, and efficient supply chain ecosystem, which can sustain large scale disruptions. In this direction, the implementation of dynamic SLAs and decentralized learning techniques is significantly enhancing the project's platform ability to respond to unexpected changes and disruptions.
Figure 1: Sequence diagram of the Federated Learning component implemented with RestAPI
