(429c) Condition-Based Sensor Health Monitoring Using Slow Feature Analysis

In the 4-th industrial revolution, biotech industry tends to apply digital technologies, e.g. Machine Learning, to make the manufacturing process more intelligent and efficient. In biomanufacturing, sensors play critical roles in collecting and transmit process information to control systems, operating personnel or upper level decision making tools. A bioreactor is normally equipped with several temperature sensors that require accurate clean steam temperature control to ensure consistent product quality and qualified sterilization of the upstream and downstream biopharmaceutical production train. Sensor failures and malfunctions are often observed and to avoid this, time-based maintenances (TbM) are performed to keep sensors in good condition[1]; wherein, maintenance schedules are following predetermined guidelines with fixed timeline which does not take the impact of different production environments to sensors’ lifecycle and sensor random failure into consideration. In contrast, condition-based maintenance (CbM) strategy [2] is an intelligent program that enables the maintenance decisions to be made based on the real-time health condition of sensors and allows the maintenance activities to be performed according to needs [3].

In this work, a condition-based maintenance (CbM) framework for real-time sensor health monitoring is proposed and implemented in a biomanufacturing process. In this CbM framework, a Slow Feature Analysis (SFA)-based approach is proposed to diagnose sensor health, detect sensor failures, i.e. slow and fast drifts, dropout, oscillation, and fluctuations. By using SFA, input signals can be decomposed into slowly and fast-varying features [4]. Different types of failures can be detected and prognosticated by monitoring different features according to the faults’ slowness in nature. The propose approach fully automates the selection of feature’s slowness for different faults and calculate a comprehensive health metric for each sensor, hence provide intelligent and flexible maintenance guides to operations in real time and greatly improve the operation efficiency and reduce OPEX. The proposed CbM strategy is deployed as part of the corporate sensor-health management system, and its efficacy is verified in a cell culture process.

References:

[1] Prajapati, A., Bechtel, J., and Ganesan, S. (2012). Condition based maintenance: a survey. Journal of Quality in Maintenance Engineering, 18(4), 384-400.

[2] Yam, R., Tse, P., Li, L., and Tu, P. (2001). Intelligent predictive decision support system for condition-based maintenance. The International Journal of Advanced Manufacturing Technology, 17(5), 383-391.

[3] Tulsyan, A., Garvin, C. and Undey, C. (2020) Condition-based sensor-health monitoring and maintenance in biomanufacturing. IFAC World Congress.

[4] Wiskott, L. and Sejnowski, T.J. (2002). Slow feature analysis: Unsupervised learning of invariances. Neural computation, 14(4), 715-770.