Intelligent vehicle health monitoring through engine data, artificial intelligence, and machine learning

Data rejection and filtration are required in this step to remove outliers and noise, to get a realistic picture of normal behaviour. The output of a health monitoring system is usually a numerical quantity or an indicator that quantifies the condition of the monitored system's component or subsystem. Different conditions can be represented by different values of such indicators. These features capture higher-level information in the sensor data. The parameters acting as condition indicators for faults are identified and monitored to detect, identify, and characterise faults by studying anomalies and trends. Diagnostic processes allow the rapid determination of specific components that need to be replaced during maintenance. Prognostic processes enable the prediction of the residual life of components by analysing trends in historical observations. A scheme capable of performing fault detection and identification has to be developed first. In case faults are identified, isolation schemes should indicate the degraded subsystem or part of the system which is affected by the fault. Finally, a set of different nudges has to be identified and assessed, regarding the more or less strong deviation from expected performance that is introduced by the fault and its progression. This assessment has to be performed either by making direct use of a generic degradation model or by employing machine learning techniques.

Once the health indicators for all characteristic faults of each critical subsystem of a vehicle have been clearly identified, artificial intelligence techniques can be employed to identify trends, gain insights from the massive volume of data, and make inferences from them. Machine learning refers to techniques designed to take in information and learn from it. These systems have the capability to evaluate and categorize received data and draw inferences from the data. A diagnostic system based on machine-learning techniques has the capability to automatically detect the best predictors of system failure. An intelligent vehicle level health management would require a robust reasoning system that could clearly distinguish between the different layers of the vehicle. Since the various subsystems of an APC have unique health indicators, suitable algorithms must be chosen for data processing, feature selection, and extraction. The choice of algorithm would be based on the requirements of the system and the data being processed for insights.