Internet of Things (IoT) and data mining techniques have laid the foundation for the next generation of smart and secure manufacturing systems where big data are leveraged to extract useful information about the manufacturing processes and further help optimize decisions. The threat of data breach exists especially for nonpersonal, yet sensitive data, which are pertinent to every aspect of manufacturing. Data breach and privacy leakages can significantly impede the manufacturer's business and lead to damaging a company's reputation. With a comprehensive case study in the manufacturing setting, we show that adversaries can utilize accessible shop floor predictive models and other available background information to make inferences about sensitive attributes that were used as inputs to the original model and use that information for their own purposes. From this view, this article presents a privacy-preserving data mining framework to build a smart and secure manufacturing system. First, we introduce differential privacy (DP), an emerging approach to preserve the individual's privacy in the data mining process. Second, we present a privacy-preserving system where DP mechanisms and queries are enforced to obtain differentially private results. Third, we propose to optimize the selection of DP mechanisms and privacy parameters by balancing the model utility and the robustness to attack. Further, we evaluate and validate the proposed privacy-preserving data mining framework with a real-world case study on the modeling of cutting power consumption in computer numerical control turning processes. Experimental results show that the performance gain, i.e., the trade-off between model utility and the robustness to attack, is improved from the nonprivate model by 5.6, 9.4, and 13.1 % for privacy-preserving Laplace, Gaussian, and sensitive mechanisms, respectively. This article is among the first to investigate and present a privacy-preserving data mining framework for smart manufacturing. The proposed methodology shows great potential to be generally applicable in industry for data-enabled smart and sustainable manufacturing.
All Science Journal Classification (ASJC) codes
- Control and Systems Engineering
- Computer Science Applications
- Industrial and Manufacturing Engineering