Data Anonymization for the Industrie 4.0

With initiatives such as the ``Industrie 4.0'' pushing for industrial digitalization, new opportunities are emerging for manufacturers to use the big data generated and collected throughout the entire lifetime of the factories. This data can be analyzed to improve the production process and create smarter factories. Analytics services are typically externalized on the cloud. This allows manufacturers to rely on external resources and infrastructures for computing models able to understand specific phenomena and predict future scenarios. While data analysis can produce results that might lead to significant improvements in the production process, the disclosure of intellectual property is a serious corporate privacy risk. Data anonymization represents a solution that allows data analysts to extract useful information from the data without risking to disclose corporate secrets. This thesis explores and categorizes the different techniques for data anonymization existing in the literature and their application on industrial data. With many privacy models and techniques available, selecting the right one is not a trivial problem. To obtain the best results in terms of privacy and quality, we propose a tool to collect the anonymization and analytics requirements, in order to tailor the anonymization to the analytics task. We then explore the application of data anonymization techniques on two use-cases for the outsourcing of data analytics tasks coming from the process industries. In the first use-case, we study the case of pattern mining on event-log data and explore the application and performance in terms of quality and privacy of a state-of-the-art framework. Our contribution moreover includes a distributed extension the original algorithm, which provides additional corporate privacy protection and better mining performance. In the second use-case, we study the task of reliability analysis on equipment data. We investigate the application of two different anonymization techniques based on encryption and randomization principles. We then conclude with a discussion on the current state of anonymization on industrial data, and discuss the future outlook for further research. The work has been developed within the ABB Corporate Research center in Ladenburg, Germany