Sensor Failure Detection, Identification, and Accommodation Using Fully Connected Cascade Neural Network

Hussain, Saed; Mokhtar, Maizura; Howe, Joe M.

Abstract

Modern control systems rely heavily on their sensors for reliable operation. Failure of a sensor could destabilize the system, which could have serious consequences to the system's operations. Therefore, there is a need to detect and accommodate such failures, particularly if the system in question is of a safety critical application. In this paper, a sensor failure detection, identification, and accommodation (SFDIA) scheme is presented. This scheme is based on the fully connected cascade (FCC) neural network (NN) architecture. The NN is trained using the neuron by neuron learning algorithm. This NN architecture is chosen because of its efficiency in terms of the number of neurons and the number of inputs required to solve a problem. The SFDIA scheme considers failures in pitch, roll, and yaw rate gyro sensors of an aircraft. A total of 105 experiments were conducted; out of which, only one went undetected. The SFDIA scheme presented here is efficient, compact, and computationally less expensive, in comparison to schemes using, for example, the popular multilayer perceptron NN. These benefits are inherited from the FCC NN architecture.

Citation

Hussain, S., Mokhtar, M., & Howe, J. M. (2015). Sensor Failure Detection, Identification, and Accommodation Using Fully Connected Cascade Neural Network. IEEE Transactions on Industrial Electronics, 62(3), 1683-1692. https://doi.org/10.1109/tie.2014.2361600

• Journal Article Type: Article
• Acceptance Date: Sep 1, 2014
• Online Publication Date: Oct 8, 2014
• Publication Date: 2015-03
• Deposit Date: Apr 25, 2017
• Journal: IEEE Transactions on Industrial Electronics
• Print ISSN: 0278-0046
• Electronic ISSN: 1557-9948
• Publisher: Institute of Electrical and Electronics Engineers
• Peer Reviewed: Peer Reviewed
• Volume: 62
• Issue: 3
• Pages: 1683-1692
• DOI: https://doi.org/10.1109/tie.2014.2361600
• Keywords: sensors, computerised instrumentation, failure analysis, fault diagnosis, gyroscopes, learning (artificial intelligence), neural nets
• Public URL: http://researchrepository.napier.ac.uk/Output/832571