Skip to Main Content
 

Global Search Box

 
 
 
 

Files

File List


41570.pdf (7.84 MB)

ETD Abstract Container

Abstract Header

Domain-based Collaborative Learning for Enhanced Health Management of Distributed Industrial Assets

Pandhare, Vibhor
http://orcid.org/0000-0002-1956-6165
http://rave.ohiolink.edu/etdc/view?acc_num=ucin1636389110876618

Abstract Details

2021, PhD, University of Cincinnati, Engineering and Applied Science: Mechanical Engineering.
The advent of technology paradigms such as Artificial Intelligence, Internet of Things, Cyber-physical Systems, and initiatives like Industry 4.0 direct to a future where intelligence is endowed to every value-adding entity across enterprises. Building on this foundation, the real fruits of sustainable, synergistic, and service-based profit can only be reaped through collaboration between these entities. One of the most significant outcomes of collaborative operation is servitization: selling of services instead of selling the assets that provide the service. For servitization in the manufacturing scenario, the upkeep of the assets becomes even more important as downtime directly transfers to the revenue lost. This means the revenue generated is directly proportional to asset availability. This requires effective monitoring of the asset fleets which are now distributed across organizations. Thus, the dissertation addresses four major challenges of collaborative learning in the domain of industrial application for enhanced health management of distributed assets: collaborative distributed model training, preserving data privacy, ensuring computation security, and addressing machine-to-machine variations for knowledge transfer. Federated Expectation-Maximization algorithm is proposed for unsupervised baseline model training preserving data privacy. A regression-based anomaly avoidance algorithm while evaluating model convergence is proposed to ensure the correctness of the trained global model in the presence of a malicious collaborator. A systematic methodology for collaborative prognostics based on the federated global baseline model is proposed. A convolutional neural network-based domain adaptation method is proposed that minimizes the maximum mean discrepancy of high-level representations between domains and exploits novel parallel data to attain class-level alignment for addressing variations in data collection configuration across machines. The usefulness and performance of the proposed methods are validated using five case studies on three datasets - a simulated dataset, the NASA turbofan engine dataset, and IMS ball screw dataset. The proposed federated approach with parameter sharing is shown to perform at par with the traditional approach with data sharing on the simulated dataset as well as the turbofan engine dataset. The proposed federated model further demonstrates improved robustness of predictions made collaboratively keeping the data private compared to local predictions for the turbofan engine dataset. The methodology facilitates effective learning of asset health conditions for data-scarce organizations by collaborating with other organizations to preserve data privacy. This is most suitable for a servitization model for Overall Equipment Manufacturers who sell to multiple organizations. The work also shows the impact of malicious behavior by a participant on global model training performance and the effectiveness of the proposed attack avoidance method to ensure model correctness. The proposed domain adaptation method achieved a mean testing accuracy of 98.25% upon validation on 33 transfer tasks designed across five accelerometer locations on a ball screw testbed depicting nine health conditions through variations in preload levels and backlash. This convenience of transferability of the diagnostic model between sensor locations can go a long way in robust and reliable condition monitoring of critical assets. Overall, the proposed work can expedite the adoption of predictive maintenance by making local expertise in analytics and platform technologies unnecessary.
Jay Lee, Ph.D. (Committee Chair)
Jay Kim, Ph.D. (Committee Member)
Boyang Wang (Committee Member)
David Siegel, Ph.D. (Committee Member)
Manish Kumar, Ph.D. (Committee Member)
127 p.
Mechanical Engineering
Collaborative Learning; Federated Learning; Domain Adaptation; Expectation Maximization; Data Privacy; Prognostics

Recommended Citations

Citations

  • Pandhare, V. (2021). Domain-based Collaborative Learning for Enhanced Health Management of Distributed Industrial Assets [Doctoral dissertation, University of Cincinnati]. OhioLINK Electronic Theses and Dissertations Center. http://rave.ohiolink.edu/etdc/view?acc_num=ucin1636389110876618

    APA Style (7th edition)

  • Pandhare, Vibhor. Domain-based Collaborative Learning for Enhanced Health Management of Distributed Industrial Assets. 2021. University of Cincinnati, Doctoral dissertation. OhioLINK Electronic Theses and Dissertations Center, http://rave.ohiolink.edu/etdc/view?acc_num=ucin1636389110876618.

    MLA Style (8th edition)

  • Pandhare, Vibhor. "Domain-based Collaborative Learning for Enhanced Health Management of Distributed Industrial Assets." Doctoral dissertation, University of Cincinnati, 2021. http://rave.ohiolink.edu/etdc/view?acc_num=ucin1636389110876618

    Chicago Manual of Style (17th edition)

ucin1636389110876618
159
© 2021, all rights reserved.
This open access ETD is published by University of Cincinnati and OhioLINK.