A Risk Evaluation Method Before Using a Black-box Model Considering Sensor Failure.

When developing an algorithm, data from sensors is often assumed to be perfect. However, sensors can fail, and a robustness function test alone may not be enough to convince customers to use a black-box in a critical durable system, because it doesn't take into account the impact of a wrong decision made by the black-box algorithm. A risk evaluation method has been proposed to assess the negative impact caused by sensor failure when using a black-box. The identification and quantification of sensor failures and risks are incorporated into the assessment process. This method has been proven useful through a case study of an access gate, where it was used to rank the importance of each sensor for the operator.