DESIGN-IT

Low-code programming enables the design, configuration, deployment, and operation of IT solutions without requiring specific expertise in software. The project aims to provide non-expert users with a graphical interface to compose solutions based on multi-paradigm Artificial Intelligence modules (e.g., machine learning, simulation, mathematical optimization) using a ‘drag-and-drop’ approach. This interface will allow users to:

• Select a deployment mode (e.g., a server from a cloud provider).
• Choose an existing Digital Twin or generate one using wizards accessible to domain experts rather than software experts.
• Select a series of Artificial Intelligence modules to include in the computation pipeline.
• Specify a source and destination for the data necessary for the application’s functioning.
• Configure the inputs and outputs of each module and design a multi-paradigm computation pipeline, defining the inputs of some modules as outputs of others.
• Parameterize the algorithmic modules and set data ingestion frequencies and module execution at runtime.
• Deploy and run the defined computation pipeline, with the ability to visualize certain predefined Key Performance Indicators (KPIs).

A Digital Twin is a virtual model designed to precisely mirror a physical object or process. These techniques are typically specialized either in processes (e.g., production processes, supply chains) or in products (e.g., CAD modeling of products, mathematical-physical analysis of material behavior). DESIGN-IT aims to integrate within a single framework the specificities of discrete event simulation techniques, kinematic simulation, mathematical-physical analysis of material behavior (i.e., FEM method), and machine learning. The goal is to generate multi-level Digital Twins that cover the entire lifecycle of a product/machine/process.

Supporting humans throughout the lifecycle of a product/process involves continuously updating Digital Twins from diverse data sources, with varying granularity and frequencies. For instance, during the design phase, data sources will primarily rely on domain expert experience. However, during the operation phase of the real counterpart, the Digital Twin must be frequently and automatically updated based on data sources such as sensors present on products or machinery.

The project aims to develop Artificial Intelligence modules available within the Decision Science Design Platform. These algorithmic modules will be based on machine learning and mathematical optimization techniques that prioritize explainability, trustworthiness, fairness, and robustness criteria. These criteria are increasingly crucial as Artificial Intelligence becomes more pervasive.

The AI algorithmic services within the platform will, for instance:

• Optimize the design of products/machines/processes (e.g., layout and production process, supply chain) from a risk-based perspective, considering the intrinsic variability of supporting data. This will be achieved through the interaction between stochastic optimization algorithms and Digital Twins.
• Predict future trends and identify anomalies in advance based on machine learning techniques (both supervised and unsupervised) and historical series analysis/forecasting.
• Optimize the operation of products/machines/processes throughout their lifecycle using predictive algorithms like predictive/prescriptive maintenance, production planning, and scheduling.
• Expedite computation times and reduce hardware consumption by gradually substituting complex and time-consuming simulation models with adequately trained machine learning models using generated data.