Tampering Detection for Automotive Exhaust Aftertreatment Systems using Long Short-Term Memory Predictive Networks

The act of tampering can be defined as a single event ranging from actions such as resetting the reading of an odometer to more advanced and long-term actions such as manipulation of the vehicle's emission control systems. Tampering, however, requires certain interventions and changes to be made on the vehicle. Recently, the sophistication of certain vehicle sub-systems such as the emission control system, have also increased the sophistication of the tampering devices. Nowadays, tampering involves not only physical changes to certain automotive sub-systems, but also the manipulation of communication signals in order to hide the presence of tampering devices. This paper presents a detection method addressing tampering of the Automotive Exhaust Aftertreatment Systems. The proposed approach leverages Long Short-Term Memory predictive networks as detection models together with Cumulative Sum control charts. The proposed detectors were validated on datasets produced by a state-of-the-art aftertreatment simulation model of a heavy-duty vehicle. The datasets encompass diverse driving scenarios alongside known and unknown (e.g., possible future) tampering methods.