Homomorphic Encryption as a secure PHM outsourcing solution for small and medium manufacturing enterprise
Introduction
Prognostics and Health Management (PHM) systems are designed to detect whether a component of an industrial machine has diverted from its normal operating condition [1]. As an important element of managing system health, PHM techniques provide predictive insights to help prevent catastrophic machine failures [2]. Advancement in information analytics and the increased availability of data due to the global adoption of Internet of Things (IoT) in factories [3,4] has enabled significant progress in the development of PHM technologies [4,5]. Major leaders in advanced manufacturing, such as General Electric have interconnected their physical factory floors and cyber computational spaces, creating Cyber-Physical System (CPS) [6,7], to enhance PHM capabilities. The establishment and use of advanced PHM technologies are disrupting many factories in the Industry 4.0 era [8,9].
Small and medium enterprises (SMEs) lag behind in the use of advanced PHM technologies often due to lack of skills and resources required for PHM-related data analytics [[10], [11], [12]]. Accordingly, cloud-based PHM that offers SMEs a possibility to receive PHM as a service in the cloud attracted great attention in the literature [[12], [13], [14]]. Outsourced scenarios of PHM-analytics tasks include determining remaining useful life of aircraft engines fleet [15], analyzing failure modes of motors [16], and recognizing health states of sawing machines [13]. However, one limiting factor to such PHM outsourcing is the data owner’s concern with respect to data privacy and security [11,17,18]. Without the development of privacy-preserving analytics technologies, the idea of PHM servicing remains incomplete [11,16,17].
Privacy-preserving data mining is the method of extracting information from data owner’s private data without revealing the data [19,20]. One way to achieve such privacy-preserving objective is to use randomization techniques from signal processing and statistics [21,22] that guarantee manufacturer’s data privacy by perturbing the data in a predefined manner [23]. Such idea has been implemented in a number of smart manufacturing systems, including cyber physical systems [24], participatory sensing [25], and smart grids [26] in the form of differential privacy and k-anonymity [27,28]. However, in such randomization techniques, there exists a tradeoff between data privacy and data accuracy, since privacy is achieved by blurring information of the original data via noise addition or de-identification [29]. For PHM analyses on CPS that require precision, e.g. machine failure diagnosis, such sacrifice in data accuracy can pose danger [24].
Cryptographic methodologies offer an alternative way to perform privacy-preserving data mining, but without the loss in accuracy [29]. A popular method is Secure Multiparty Computation (MPC), a protocol that involves multiple parties to do joint computation on data to obtain useful information while keeping the data private [30]. In the manufacturing setting, MPC protocol was demonstrated in a variety of applications including semiconductor management [31], internet of vehicles (IoV) protocol [32], and supply chain management [33]. However, the basic premise of using the MPC protocol is that all participants must participate in cryptographic communications during the entire round of online interaction, which makes the adoption challenging to SMEs who lack cryptographic knowledge.
Homomorphic Encryption (HE) is an encryption technique that allows a server to do direct computations on client’s encrypted data [34]. Under HE techniques, data is encrypted in a way such that the results of “direct” computation on the encrypted data are identical to the results obtained from computation on the original unencrypted data. By “direct”, we mean that the data is never decrypted and that analytics are performed directly on the encrypted data. Because the computation is done solely by the server, HE eliminates the need for cryptographic communication between a client and a server. Moreover, while MPC makes an assumption that participants do not collude each other, HE does not require such assumption; it allows situations that involve clients who cannot trust the server. In addition, it should be emphasized that just like the MPC protocol, HE does not make a compromise in accuracy while achieving privacy.
From such advantages, HE has demonstrated its wide applicability as a privacy-preserving computational tool in medicine [35], bioinformatics [36] and finance [37], of which the issue of privacy is obvious. Recent literatures in manufacturing also outlined some scenarios of applying HE onto IoT architectures [[38], [39], [40]], CPS architectures [41], and distributed analytics frameworks [42] as well. However, to the best of our knowledge, HE as a secure PHM outsourcing solution for SME’s machine maintenance tasks has not been showcased in the literature to date.
In this paper, we demonstrate how HE can enable SME’s secure outsourcing of health diagnostics on a manufacturing machine with high precision. We first outline the two-party collaborative framework involving a client (SMEs) and a server (PHM analyst) for a secure PHM outsourcing. Next, we introduce a HE frequency analysis algorithm (H-FFT-C) to demonstrate one technique under the proposed scenario. Then, we apply this algorithm to a lab-scale system, a Fiber Extrusion Device (FrED) [43], as an example of the privacy-preserving PHM outsourcing scenario proposed for SMEs. Finally, the extensibility and key limitations of our proposed approach is outlined in our discussion. We envision that our proposed framework and algorithmic methodology will advance the use of Homomorphic Encryption for outsourcing of complex PHM-analytics.
Section snippets
Background on homomorphic encryption
Cryptographers study and use hard problems. To build the most secure firewall around data and network traffics, organizations have no choice but to install “the hardest problems” into their security system to prevent adversarial attacks. One of the state-of-the-art hardness assumptions used in the modern cryptography is so-called Learning With Errors (LWE) that was introduced by Regev (2005) [44]. In his work, it is proven that the LWE problem is as hard to solve as several worst-case lattice
Demonstration on a manufacturing machine
In this section, we apply our H-FFT-C algorithm on a lab-scale fiber extrusion system called FrED [43] to showcase the scenario on privacy-preserving PHM outsourcing for SMEs. Here, we assume that an enterprise purchased FrED and extrudes fibers to make a product. We show how the FrED user can receive an accurate maintenance service from a PHM analyst while keeping all data private, by engaging in the two-party collaborative framework described earlier.
Extensibility
The methodology proposed in this study can capture additional characteristics of FrED to diagnose machine failure in a privacy-preserving manner. First example is the direct detection of component obsolescence via H-C algorithm. Fig. 7(a) plots the evolutions of fiber diameter measured from a laser micrometer. Note that FrED with an old spool generates a product with oversized diameter. By directly applying the H-C algorithm with a threshold value near 0.35 mm, the oversized product can be
Conclusion
In this paper, we introduced a two-party collaborative framework and methodology for a secure outsourcing of prognostics and health management (PHM) on manufacturing machines for small and medium enterprises (SMEs). In our approach, a cryptographic method called Homomorphic Encryption (HE) is used to enable privacy-preserving computation on client’s private data. The advantage of using HE in such privacy-preserving data mining scenario comes in twofold. First, as opposed to randomization-based
Declaration of Competing Interest
The authors report no declarations of interest.
Acknowledgements
This work was supported in part by Liberty Mutual Insurance, Boston, MA as part of a collaboration with The MIT Quest for Intelligence. We thank the homomorphic encryption team at the Microsoft Research for helpful discussions on the SEAL library [56].
References (56)
- et al.
Service innovation and smart analytics for industry 4.0 and big data environment
Procedia Cirp
(2014) - et al.
Current status and advancement of cyber-physical systems in manufacturing
J Manuf Syst
(2015) - et al.
A cyber-physical systems architecture for industry 4.0-based manufacturing systems
Manuf Lett
(2015) - et al.
Cyber-physical systems architectures for industrial internet of things applications in Industry 4.0: a literature review
J Manuf Syst
(2021) - et al.
Recent advances and trends in predictive manufacturing systems in big data environment
Manuf Lett
(2013) - et al.
The WEAR methodology for prognostics and health management implementation in manufacturing
J Manuf Syst
(2017) - et al.
A critical review of smart manufacturing & Industry 4.0 maturity models: implications for small and medium-sized enterprises (SMEs)
J Manuf Syst
(2018) - et al.
Artificial intelligence and internet of things in small and medium-sized enterprises: a survey
J Manuf Syst
(2021) - et al.
Security of smart manufacturing systems
J Manuf Syst
(2018) - et al.
Industrial data management strategy towards an SME-oriented PHM
J Manuf Syst
(2020)