Online Fault Detection of Rear Stroke Suspension Sensor in Motorcycle

This paper describes the implementation and experimental verification of an instrument fault detection (IFD) scheme for stroke sensors. In the scheme, a nonlinear autoregressive with exogenous inputs (NARX) neural network works as a soft sensor for the generation of residuals, while a rule-based decision maker provides the fault detection and classification. The scheme was thought to be implemented in the firmware of central units for the control of semiactive suspension systems for motorbikes. Execution times compatible with the real-time application constraints are reached through straightforward programing rules and suitable code optimization. These times were obtained with machine parameter values (such as clock frequency and absorbed current) far below the upper limit of the range, thanks to the adoption of a programing methodology specially designed for the real-time implementation of diagnostic schemes on general-purpose microcontrollers (MCUs). The diagnostic performance of the scheme was verified through an experimental campaign, carried out on a motorcycle featured with electronic controlled semiactive suspensions. Promptness and reliability in detecting the most common faults of the rear stroke sensor resulted fully compatible with actual applications' expected values.