Decision Forest for Root Cause Analysis of Intermittent Faults

Intermittent failures can be problematic in electronic control units (ECUs) such as engine/transmission control modules. When an ECU exhibits an internal performance fault, the ECU may malfunction, while the fault condition is active, and later, it may once again give correct results when conditions change. Due to highly unpredictable nature of intermittent faults, it can be extremely difficult to diagnose them. Therefore, there is a need to enhance the fault diagnosis of intermittent faults in ECUs. In this paper, we propose an off-board, data-driven approach that can assist diagnostic engineers to investigate intermittent faults using fleet-wide field failure data. The field failure data may include a large number of intermittent faults and concomitant operating parameters (e.g., vehicle speed, engine speed, control module voltage, powertrain relay voltage, etc.) recorded at the time when the faults occurred. We describe a decision forest method to identify a reduced set of informative operating parameters, i.e., features that separate or characterize the operating conditions of the intermittent fault from baseline, i.e., classes in feature selection space. A web-based application has been developed to assist the diagnostic engineers. We demonstrate the capabilities of our method using three case studies for an automobile test fleet.