Andrey Kurzyukov

Andrey Kurzyukov

Master's Thesis

Modeling Mixed-Type Time Series Data for Machinery Health Prognostics

Advisors

An Nguyen (M.Sc.), Leo Schwinn (M.Sc.), Dr. Dario Zanca, Dr. Michael Moser (Siemens Healthineers),  Prof. Dr. Björn Eskofier

Duration

09 / 2021 – 02 / 2022

Abstract

Predicting when a machine is likely to fail plays a crucial role in any business involving equipment because it helps plan maintenance in advance [2]. Condition-based maintenance (CBM) monitors the health status of machinery and decides the need for maintenance based on historical data [1]. Machinery health prognostics (MHP) is a subtask of CBM, which focuses on predicting the future machine health state. The future state of a machine can, for example, be modeled by estimating the remaining useful lifetime (RUL).

Most related works describe how to perform MHP based on evenly sampled and high-frequency sensory data (e.g., accelerometer data) [2, 3, 4], while others – based on event data [5, 6]. Additionally, few researchers fuse operational and event data to predict the next event type and timestamp [7, 8]. However, limited research is done to model MHP on mixed-type time series data a previous thesis investigated.

This thesis aims to design, implement, and evaluate an approach for MHP based on mixedtype time series data using machine learning (ML) models. It is an open question if a model can be designed and implemented, which yields practical results for potential deployment. A previous thesis showed that standard approaches do not work (e.g., making predictions based on small windows with an i.i.d. assumption). A major medical device company provides data based on a critical component from a large fleet of systems installed at clinical facilities. The dataset includes a few thousand life cycles of these components. The data comprises irregularly sampled sensor measurements (per machine operation), event logs, service notifications, and usage statistics. In summary, the available data imposes the following challenges:

• Different data sources need to be integrated
• Long and varying time series to model (e.g., several 100k sensor measurements and events per machine component life cycle)
• Varying utilization of CT scanners between healthcare facilities
• Collected data is noisy since it is collected from real healthcare facilities

 

References:

[1] Lei Y., et al.: Machinery health prognostics: A systematic review from data acquisition to RUL prediction. Mechanical systems and signal processing. – 2018.
[2] Mathew V., et al.: Prediction of remaining useful lifetime (RUL) of turbofan engine using machine learning. 2017 IEEE International Conference on Circuits and Systems (ICCS). – IEEE, 2017.
[3] Li X., et al.: Remaining useful life estimation in prognostics using deep convolution neural networks. Reliability Engineering I& System Safety. – 2018.
[4] Nguyen K. T. P., et al.: PA new dynamic predictive maintenance framework using deep learning for failure prognostics. Reliability Engineering I& System Safety. – 2019.
[5] Sipos R., et al.: Log-based predictive maintenance. Proceedings of the 20th ACM SIGKDD international conference on knowledge discovery and data mining. – 2014.
[6] Calabrese M., et al.: SOPHIA: An event-based IoT and machine learning architecture for predictive maintenance in industry 4.0. Information. – 2020.
[7] Xiao S., et al.: Learning time series associated event sequences with recurrent point process networks. IEEE transactions on neural networks and learning systems. – 2019.
[8] Xiao S., et al.: Modeling the intensity function of point process via recurrent neural networks. Proceedings of the AAAI Conference on Artificial Intelligence. – 2017.
[9] Ribeiro et al.: Why should i trust you?: Explaining the predictions of any classifier. Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. 2016.
[10] Lundberg, Scott M., and Su-In Lee.: A unified approach to interpreting model predictions. Advances in Neural Information Processing Systems. 2017.
[11] Lai, Guokun, et al.: Modeling long-and short-term temporal patterns with deep neural networks. The 41st International