Secure Future Digital Manufacturing – Privacy-preserved IoT services


Future deployments in Industry 4.0/5.0 manufacturing scenarios will use Internet-of-Things (IoT) platforms as a service. Consequently, IoT platforms can give key insights to on-site workers tasks and usage of devices, machines and instruments, along with their movements.

Then how can useful information, insights and performance analyses be made identifying activities — while protecting privacy and confidentiality? (I) Protecting Privacy — to respect personal sensitivities of workers, according to particular workers’ movements across time and their precise activities. (II) Protecting Confidentiality — to manage for commercial sensitivities, for a service-provider to multiple manufacturers, with confidentiality around precise usage

Inference about activities and usage is a vital purpose of IoT deployments in Industry 4.0/5.0. Thus, there is a need to know what is happening with IoT Sensor Tags, where they are deployed, and how they are moving.


Our project has sought an answer, developing provably-private machine learning, AI, for IoT deployments for digital manufacturing.

Thus, our team has tested this at CSIRO’s labs of the future with deployments at QCAT (Pullenvale, Queensland, Australia) and Clayton (Victoria, Australia). Hence we have used the Embedded Intelligence Platform (EIP), an Internet-of-Things (IoT) solution developed at CSIRO Data61, with one particular focus of its implementation being digital manufacturing. Key facets of such EIP IoT deployment – with the equipment and potential personnel monitoring it provides in a manufacturing scenario – are privacy/confidentiality-preserving solutions that can still make useful inferences, while providing accessible information from the collected data. We have provided three privacy-preserving machine-learning (ML) solutions based on real-life data, from the deployments from the CSIRO APAIR project on Future Digital Manufacturing (FDM) – the project by which this work has been enabled and funded. The solutions are:

(i). ML/AI Classification of the EIP IoT tags (dynamic and static) and their use.

(ii). ML/AI has been distributed to apply in locational clusters of tags, such that the classification can be deployed behind EIP gateways.

(iii). Time-series prediction to predict key parameters in near real-time across several time steps for six key attributes of EIP sensor-tag data

Output of Privacy-preserved — Classification and Time-Series Prediction — at CSIRO QCAT, including map of active EIP IoT Tags


Chia, S. Y., Xu, X., Ding, M., Smith, D., Paik, H. Y., & Zhu, L. (2023, March). A Selection Model of Privacy Patterns. In 2023 IEEE 20th International Conference on Software Architecture (ICSA) (pp. 1-11). IEEE.