Periodic Reporting for period 2 - ASSISTANT (leArning and robuSt deciSIon SupporT systems for agile mANufacTuring environments)
Reporting period: 2022-05-01 to 2023-10-31
• An intelligent digital twin for process design or redesign: What resources, tools, skills do we need, and how can we organize them?
• An intelligent digital twin for production planning: How much do we produce each week? ASSISTANT will integrate machine learning techniques (with optimization) to process and exploit small and large datasets.
• An intelligent digital twin for scheduling: Assignment of products to the machine and the order in which operations are performed. ASSISTANT will provide a model acquisition tool that enriches basic scheduling models with learned constraints.
• A data fabric that collects data from IOT devices, machines, operators, and existing software. Data will be cleaned and stored in dynamic knowledge bases.
• Tools for safe decision actuation, and control. This includes the development of flexible cognition methods and resource management for collision-free planning and adaptive motion planning on the shop floor. These actuation tools will allow one to react in a timely manner to unforeseen events due to autonomous, real-time, and optimization-based methods improving time, costs, and safety conditions, among other economical, environmental or social considerations.
• ASSISTANT will provide AI Management and Assessment Plan on Ethics to guide ASSISTANT partners in the development and deployment of responsible AI solutions.
Once the requirements and specifications were defined, and a first version of the architecture defined, the project moved into the development phase. At this stage of the project, the main achievements are:
• 21 deliverables were submitted out of a total of 51 deliverables
• 4 milestones have been achieved. These milestones relate to (a) the kickoff of the project, (b) the first report on the architecture of the project, (c) the development of the Graphical User Interfaces and, (d) the development of the first prototypes on the digital twins on process planning, production planning and scheduling, and real-time control and actuation. In terms of dissemination, the project has 26 scientific articles in total, of which 9 published, 9 accepted and 8 under review. Regarding communication, the project has set up a website and a web page with more than 2,000 unique visitors, social media, a video explaining the concepts of the project and the expected results with more than 12,000 views on all channels for communication and dissemination of the ASSISTANT project. Seven press releases were published by the whole consortium, 8 newsletters including 4 special newsletters at the project level and 7 interviews where ASSISTANT concepts and expected results were communicated. ASSISTANT attended 8 workshops related to Ecosystem Building Activities Targeting EU Projects, EU Initiatives, Researchers, and Industry.
• Statistical machine learning techniques to produce simulated data in digital twins and to estimate parameters and cost functions in decision aid system.
• Bayesian network and moment matchings to acquire the probability distribution of unknown parameters
• Reasoning under uncertainty for robust and flexible manufacturing design and operations.
• Machine learning for the acquisition of high-level understandable prescriptive analytic models based on historical data and/or simulation data.
• Use generative design in synergy with constraint programming to help the user exploring alternatives for reconfiguring a production system.
• Programming abstractions for data streams to synthesize safe code for capturing qualitative and quantitative patterns on data streams in the context of IoT applications for real-time control in the shop floor.
As for expected result until the end of the project, ASSISTANT aims to provide intelligent twins for the design and operation of agile production systems. A process planning intelligent twin helps the design of production systems with a high level of flexibility while maintaining product and process quality. The production planning and scheduling twins deal with various sources of uncertainties encountered in production systems, and the real-time twin helps to react in a timely manner to unforeseen events. ASSISTANT will contribute to six impacts described below:
(1) Products and services usable in a wide range of manufacturing processes leading to agile production processes and improved quality of products and processes.
(2) Humans working together with AI systems in optimal complementarity.
(3) ASSISTANT leverages on AI to positively impact employment and quality of jobs.
(4) ASSISTANT will support the production of a legal and ethical framework for AI at the European level.
(5) ASSISTANT will contribute to breaking down silos of research in manufacturing by putting AI research community to the benefit of the manufacturing community.
(6) ASSISTANT decision support system tools consider sustainable metrics, allowing to improve the economic, environmental, and/or social perspectives of manufacturing companies.