The challenges faced by the manufacturing sector in high-wage economies like Australia are well-documented. Manufacturers must contend with rising minimum wages, stagnant productivity, and an increasing demand to add value to differentiate their products in a global market where less advanced nations are quickly catching up with traditional technologies.
Authors – Dayalan Gunasegaram and Christian Ruberg | CSIRO Manufacturing Simon Dunstall | CSIRO Data61
One compelling strategy for survival is the adoption of “lights-out manufacturing”—a concept where machines operate autonomously, requiring minimal to no human intervention. Similar to autonomous vehicles, these factories leverage artificial intelligence (AI) to dramatically increase productivity, transforming today’s automated facilities into tomorrow’s fully autonomous factories.
While autonomous factories reduce the demand for traditional factory workers, they create opportunities for roles such as data analysts and programmers, who enable machines to “think” by leveraging machine learning (ML). ML technologies allow machines to emulate human cognitive abilities by learning from input-output relationships in a manufacturing process. These algorithms automatically uncover complex, often hidden, correlations between process parameters (inputs) and outcomes like product quality and productivity. Armed with this knowledge, ML algorithms can optimise processes and enable closed-loop control using sensor feedback to make reactive or proactive decisions for corrective actions or feedforward control, respectively.
Why Machine Learning for Autonomous Control?
Conventional modelling and control methods, such as Model Predictive Control (MPC), rely on fixed assumptions and struggle to adapt to complex, real-world environments that are dynamic and unpredictable. Consider an autonomous car navigating chaotic, ever-changing traffic—traditional physics-based models are inadequate for such scenarios. Moreover, the computational intensity of physics-based models often renders them unsuitable for real-time decision-making, as they take too long to solve. In contrast, ML models are computationally efficient, solving problems almost instantaneously, making them ideal for adaptive control strategies.
Progress and Challenges
Significant strides have been made in applying Industry 4.0 concepts to real-world manufacturing. While ML algorithms for analysing sensor data have advanced rapidly, those for decision-making are still evolving. Researchers are actively addressing these challenges, including reducing the traditionally large data requirements for ML models, which can now perform well with limited datasets.
Symposium on AI in Manufacturing
To address challenges and opportunities, CSIRO is organising a symposium on real-time AI control for manufacturing and other industries. The event will bring together global experts in ML, manufacturing (and other) professionals, researchers, and policymakers to exchange ideas, network, and develop a roadmap for the widespread adoption of advanced digital solutions. While domains usually have unique challenges, they also share some of the hurdles with other domains. Since specific strategies have been trialled or applied in some of these domains while other domains have lagged, there is the possibility for all delegates to learn from others while also contributing perspectives from their own angle. By fostering dialogue across disciplines, the event aims to accelerate the adoption of autonomous control strategies in manufacturing and other sectors. Significantly, the symposium will cater to a diverse audience, from those new to ML to experts in creating advanced models.
For more details and to register, visit the event website: https://wp.csiro.au/rtcml25/
