On Fire, and Not in a Good Way: Predictive Quality Analytics is a Game Changer

Electric and hybrid vehicles are transforming the industry, but they’re also introducing new quality and safety challenges. As automakers push toward electrification, the complexity of systems, power electronics, and thermal management has made early high-voltage (HV) battery fault detection a strategic priority. 

Over the past five years, recalls linked to battery and fire hazards have surged. Our recent analysis, using recall data from the NHTSA (U.S.) and KBA (Germany), found nearly 250 recalls citing fire risk coupled with battery defects, impacting almost 7 million vehicles. Out of these, 54 recalls were directly tied to HV batteries, and within that group, 33 cases linked HV battery defects with the fire hazard (compared to other potential risk sources), affecting nearly one million vehicles.

These findings underscore a critical reality for modern field quality investigations: the data to detect the next recall already exists, it’s streaming from the vehicles every day. All 33 HV battery fire-related recalls could have been evidenced in abnormal connected vehicle telematics patterns, and later on in Diagnostic Trouble Codes (DTCs).

Each of these abnormal patterns is a potential early warning that, if monitored proactively, could surface risks long before drivers notice an issue or the first claim is filed. The ability to detect these anomalies pre-claim and start investigations earlier represents a game-changer for OEM after-sales quality teams. With connected vehicle data, investigations no longer need to be reactive, waiting for customer complaints; they can start as soon as trends and anomalies appear in the field.

Given the acute severity of fire incidents in EV and hybrid lithium-ion batteries, where ignition cannot be stopped once thermal runaway begins and the pack may burn for hours until all stored energy is released, it is vital to underscore why proactive detection is so critical. Thermal runaway is an internal chemical reaction that sustains itself, which means that even the most advanced fire retardants have limited impact once the process starts. 

The industry’s response reflects how serious this risk has become. A number of manufacturers have introduced specialized access points that allow emergency teams to inject water directly into the battery pack and cool the cells from within, providing a practical way to slow the reaction during an otherwise difficult-to-manage event. By focusing on predictive pre-claim detection as a way to identify abnormalities in connected vehicle data, long before such irreversible thermal conditions develop, OEMs can materially reduce the likelihood of battery failures and reinforce safety.

The Disruptive Impact of Predictive Pre-Claim Detection, Using Connected Vehicle Data

Many of the failure signatures cited in the recalls included HV sensor temperature increase, cell min/max delta charging anomalies, Battery Management Systems (BMS) related DTCs, and MIL ON (dashboard indicators), which already existed within connected vehicle telemetry and diagnostic data. With the right data and ML and GenAI infrastructure, these could have served as early-warning indicators of an emerging HV battery malfunction.

Upstream’s Proactive Quality Detection (PQD) platform is designed to help after-sales teams to do exactly that. PQD continuously ingests large-scale connected vehicle data and uses ML-powered anomaly detection and digital signatures to identify subtle field deviations in system-specific behavior, long before they manifest as claims.

In similar cases, PQD has been proven to:

• Surface fleet-level anomalies in HV battery
• Correlate those signals with emerging BMS DTCs and investigate individual VINs remotely, without contacting their drivers
• Flag early risk clusters for monitoring and investigation of safety issues

This kind of predictive insight enables field engineering and after-sales quality teams to move to proactive prevention.

Expediting Investigation and Countermeasure Validation

Early investigations are instrumental in delivering cost savings. In traditional after-sales quality processes, investigations often begin after drivers notice issues or claims are filed, by which point the defect may already have propagated across thousands of vehicles. With connected vehicle data, however, the investigations can start much earlier in the quality lifecycle, at the first sign of anomalous system behavior, often before the driver experiences any symptoms.

This predictive and early detection enables field investigation teams to:

• Generate root-cause hypotheses directly from real-world vehicle data, without waiting for physical parts to be replaced, sent to the lab, or trying to replicate the issue in a test vehicle.
• Prioritize potential issues by severity, safety, and predicted impact, focusing engineering resources where they matter most.
• Accelerate validation of corrective actions, using continuous feedback from the field to confirm whether the issue has been fully remediated.

In addition, the same data-driven ML models that could have helped after-sales teams detect early signs also accelerate post-recall countermeasure validation. Once a fix is deployed, PQD AI Agents help automate TSBs and countermeasures so that the OEM’s field teams can monitor the recurrence of the known fault patterns, ensuring the effectiveness of countermeasures and preventing reoccurrence across other model lines or VIN cohorts.

This shift to proactively identifying risks in real time represents a transformative move for after-sales quality teams focusing on safety risks and reliable operations:

• Reduced safety risk and liability exposure through earlier field anomaly detection and early investigations.
• Lower warranty and recall costs by addressing issues before they scale.
• Preserved customer trust in EVs and Hybrids during a critical market adoption period.

This analysis underscores a central truth of the software-defined vehicle era: every vehicle is already generating the data needed. The question is whether that data is being harnessed. Upstream’s PQD platform, powered by purpose-built ML and LLMs and agentic AI, transforms connected vehicle data into proactive quality operations, spotting potential issues before they affect customers and enabling faster, more precise responses when they do.

The HV battery scenario reflects one of many use cases where proactive insight from connected vehicle data could have helped surface risks earlier in the cycle and might have reduced the scope and millions of dollars of later field actions.

Quick note: OEMs rely on a broad ecosystem of after-sales quality tools and processes, some of which already include predictive and pre-claim detection capabilities. The examples and use cases described here are illustrative, and OEMs typically deploy a wide range of solutions to monitor quality across the after-sales lifecycle.