The Case for AI-based Integrated Vehicle Health Management: To reduce OEM costs and the total cost of ownership (TCO) for consumers and fleets, new methods are needed to detect, predict, and diagnose vehicle health issues – quickly, efficiently, and ahead of serious part or system failures.
The increasing complexity of vehicle electronics and software is revealing an abundance of vehicle health issues. Chief among them are concerns about quality and the increasing costs of warranty claims, recalls, fleet maintenance, fuel, and fleet downtime. Added to these are the negative impacts on OEM, fleet profitability, user experience, and end-customer costs.
To reduce OEM costs and the total cost of ownership (TCO) for consumers and fleets, new methods are needed to detect, predict, and diagnose vehicle health issues – quickly, efficiently, and ahead of serious part or system failures.
Questar developers are recommending a unique approach to resolve these challenges. This involves the innovative use of highly effective and adaptable AI-based Integrated Vehicle Health Management (IVHM) systems, with automakers and fleet owners being the main beneficiaries.
The Questar team – Sasha Apartsin, Hilik Stein, Gil Reiter, Kyle Williams and Noam Moscovich – have proposed integrating AI-based Signal Integrity Monitoring (SIM) into the vehicle’s embedded electronics, where large amounts of vehicle data (big data) can be accessed in real time. SIM is a powerful and novel anomaly detection pipeline that monitors a group of vehicle signals, detects complex anomalous patterns and isolates signal faults.
The Questar team’s novel proposal of an IVHM/SIM solution, is based on an unsupervised machine learning approach. Specifically, it employs an unsupervised anomaly detection model that is based on deep learning autoencoders. Accordingly, it can detect, for example, complex anomalous patterns in engines and help identify performance degradation, predict vehicle health issues, and determine the root cause of problems.
Existing vehicle health management (VHM) solutions on the other hand, rely on diagnostics trouble codes (DTC) and limited amounts of telematics data. While these solutions can detect known failure modes and provide alerts based on predefined error codes, they are unable to detect and diagnose unforeseen failure modes that do not have hard-coded rules. Neither can they predict future vehicle health issues.
By integrating AI-based SIM into embedded vehicle electronics, automakers and fleet owners can benefit from:
One of the key features of the SIM pipeline is that it uses an autoencoder type of neural network, which can be designed very efficiently and trained in an unsupervised manner. With unsupervised learning the AI is trained using only “normal” vehicle data from a healthy vehicle. This greatly simplifies the data collection process and automates the training process, since it eliminates the need for a large number of failure examples, which are hard to obtain It also eliminates the need for to tag (i.e. label) large amounts of data, a very time-consuming process.
One of the advantages of deep learning is that it can analyze very large amounts of complex data – big data – that cannot be effectively analyzed using traditional signal processing methods.
In a case study conducted by the Questar team, multiple engine signals were collected over a 26-hour period from an Iveco Daily light commercial van with a 2.3L gasoline engine including: Engine speed, fuel rate, torque, demand torque, nominal friction torque, intake air pressure, intake manifold pressure, oil pressure, exhaust gas mass flow rate, ambient air temperature, engine intake air temperature, intake manifold temperature, exhaust temperature and coolant temperature.
The SIM pipeline was trained using the collected data, and then tested by introducing four different malfunctions that degraded the performance of the engine. The malfunctions were each introduced separately and consisted of an injector leak, an injector malfunction, an intake manifold leak, and a blocked air filter.
The tests showed a clear differentiation between the test vehicle’s normal behavior and its anomalous behavior. For each of the tested failures there was distinct separation between the health scores of a health vehicle versus those with the malfunctioning components.
It should be noted that during the tests for injector leak, intake manifold leak, and air filter blockage, the engine controller did not trigger a DTC, and the check engine lamp was not illuminated. If left uncorrected, these issues would result in increased fuel consumption, cause more pollutants to be emitted into the environment, and potentially cause significant damage to the vehicle’s engine.
However, by integrating a SIM pipeline into the embedded vehicle electronics, the degradation in the engine’s performance was detected by the SIM pipeline before any of the malfunctions became a serious health issue.
The Questar team believes that by adopting this approach – integrating artificial intelligence (AI) technology into the vehicle’s embedded electronics – automakers and fleet owners can benefit from highly effective and adaptable vehicle health management capabilities that up to now have just not been available. This will create a significant beneficial impact on Total Cost of Ownership, an issue which is top-of-mind with fleet operators today.
