AI IN MANUFACTURING

The aim of the Cyber-Physical Production System (CPPS) project was to develop intelligent and contextual decision support capabilities for Cyber-Physical Production System. This project delivered a scalable CPPS platform that covers dynamic cyber-physical modelling supported by online large-scale simulation and multi-objective, multi-level optimisation capabilities.

It also delivered contextual and intelligent decision support technologies that involve machine learning for context-aware insights and autonomous root-cause-effect analysis to support intelligent decision making. This project involved collaboration with local and overseas IHLs (Technische Universität Braunschweig, NUS, NTU) as well as Inter-RI efforts (SIMTech, NMC, ARTC).

The novelty of CPPS is to create a targeted system approach for sense & response manufacturing environment with a complete digital thread that is able to handle relationships on-the-fly to support different applications. Single source of truth, contextually accurate total visibility, reinforced by fast decision-support/making capabilities, with predictive & prescriptive capabilities and fast decision-making capabilities that is tractable and able to handle uncertainties arising in real-world problems.

The programme aims to enable robots to perform as team members alongside humans. It is developing technologies to understand tasks, procedures and human actions, as well as to learn quickly and robustly, for robots to be able to learn from humans and work safely in human environments. Collab AI is led by IHPC in collaboration with I2R, ARTC, NUS, NTU and SUTD.

The objective of this project is to develop a Predictive Maintenance (PdM) system including an on-board sensing technology that incorporates IoT and on-board signal processing for machine condition monitoring and alerting. An AI-based analytical too for machine health assessment, diagnosis, performance prediction and implementing the PdM system for 3 types of machines.

This PdM solution provides a comprehensive system consisting of sensorisation for brownfield connectivity, data routine collection, machine condition monitoring, fault diagnosis, and prognosis. Since machines in mass production are different mechanical states and behaviours, the AI models are developed based on continual learning algorithm that makes it easily adaptable to change of machine states. This novelty can help company scaling the system for multiple machines in factory.

This is an ongoing industry project collaboration with Philips.

When an end-to-end network is not optimised, it introduces unnecessary network latency and errors which causes challenges in meeting the Quality-of-Service (QoS) requirements for mission critical or delay-sensitive industrial Internet-of-Things (IIoT) applications. Hence, a deep learning-driven edge caching protocol has been developed to bring contents from a remote central server, closer to the end users.

The aim of this project is to encapsulate this protocol into a deep learning-driven edge caching-as-a-service (DECaaS). DECaaS leverages on multi-access edge computing (MEC) and deep learning methods to predict future content demands of end users before caching the predicted popular contents at MEC servers. In this way, DECaaS mitigates the problem of network latency and errors.

The novelty of DECaaS lies in the realisation of deep learning-based edge caching capabilities for reliable low-latency content delivery over both cellular and IIoT networks. DECaaS is developed for general file downloads and Open Platform Communications Unified Architecture (OPC UA) tag access for sensor data. For both use cases, DECaaS is employed to predict the demand of files and OPC UA tags before caching it at an MEC server.

DECaaS is an ongoing GAP-funded project, with interest from Singtel in advancing this technology towards commercialisation.

This research involves the development of high-throughput simulations workflow based on first-principles and CALPHAD simulations to generate structure and formation datasets of High-Entropy Alloys (HEAs). Trained on these datasets, machine-learned (ML) surrogate models are developed to predict the composition-microstructure-stability relationships of HEAs. The ML models are then utilised in optimization algorithms and Monte-Carlo based methods to identify promising compositions for experimental validation, in terms of their microstructure, properties and manufacturability.

In parallel, machine-learned interatomic interactions are also developed to enable realistic atomistic simulations of mechanical behaviour of HEAs, covering microstructural stability, short range ordering, strengthening mechanisms, elastic and yield strengths.