AWS Partner Network (APN) Blog
Excelfore Edge AI for Anomaly Detection in Connected Vehicles Using AWS
By Srini Raghavan, Partner Solutions Architect – AWS
By Jakob Gasser, Sr. Partner Development Manager – AWS
By Shrikant Acharya, CTO – Excelfore
![]() |
![]() |
Continuous improvement of connected vehicles has been the focus of the automotive Original Equipment Manufacturers (OEMs). Connected vehicles generate anomalous data from the edge. Techniques that leverage Artificial Intelligence (AI) catch anomalous data, identifying quality improvement areas in the vehicle. Autonomous driving’s future: Convenient and connected, a McKinsey report, identifies up to $300 – $400 billion in revenue resulting from autonomous driving by 2034. Pursuit of this opportunity requires continuous improvements in the reliability and functionality of automotive electronics. This is driven by the collection and use of data from large numbers of vehicles in the field.
Excelfore is a leader in Connected Mobility, an advanced tier AWS Partner, and member of recognized industry organizations such as eSync Alliance, SOAFEE-SIG, ASAM, AVNU and AUTOSAR. This blog shows how Excelfore edge-based AI solution identifies and transmits the useful data alone from the vehicle to AWS, for analysis and decisioning.
Industry Challenge
Another McKinsey report titled Unlocking the full life-cycle value from connected-car data highlights OEM efforts to provide end-to-end access to 1 to 2 terabytes of data per car each day. This is to enable continuous product and service improvements. We maintain that it is simply not reasonable to expect to upload the 1–2 terabytes of data per day. This is 2–3 orders of magnitude more than today’s top-level cellular data plans, which provide 60 or 100 Gigabytes of data per device per month. In an environment where it is not feasible to send all the vehicle data to the cloud, the challenge is in determining what data to be sent to the cloud. The challenge specifically is in identifying what data is required to provide maximum value for reliability and functionality improvements.
Solution overview
Figure 1 is an overview of Excelfore’s AI based approach for efficient data collection from the vehicle to AWS.
Figure 1: Overview of AI based approach for efficient data collection
The following five-step process shows how to collect, filter, analyze, and use the vehicle data for AI-powered anomaly detection, creating a comprehensive closed-loop system for automotive intelligence.
- eSync Data Pipeline: eSync is a standard for building a data pipeline reaching edge devices in vehicles, facilitating i) Over-the-Air (OTA) updates, ii) data gathering and iii) transmission to the cloud. For more details on the eSync-compliant data pipeline, please read our prior blog: Building a Scalable Standardized Pipeline for Automotive OTA on AWS.
- eDatX Data Filtration: eDatX is an Excelfore embedded software component that selectively aggregates and uploads data to AWS. It reduces data by 99.9%, which translates to reducing gigabytes to terabytes, before transmission.
- eDatX Data Visualization and Analysis: eDatX AWS component stores and visualizes the data collected from the vehicle fleet to analyze the performance of key automotive systems.
- Cloud-based AI Modelling of fleet-scale data: Excelfore AI Engine on AWS utilizes the collected data from eDatX and trains a model to recognize normal data patterns.
- Edge-based AI Anomaly Detection: The trained AI model is sent over-the-air to the vehicles. An edge AI engine uses this model to identify anomalous data patterns prior to filtration of the vehicle data. It triggers uploading of additional detailed data needed for detailed analysis.
Solution Architecture
The solution operates across two distinct layers: the vehicle layer, which includes components running in the vehicle, and the AWS Cloud layer, which encompasses services and resources deployed on AWS.
In-vehicle
Figure 2 describes the in-vehicle components as it pertains to the data collection capabilities. The vehicle exchanges data with AWS using Messaging Queuing Telemetry Transport (MQTT) and HTTPs protocols. The vehicle continuously gathers the data and routes to the eDatX Service, where it undergoes configurable processing.
Figure 2: In-vehicle solution architecture for eSync with the Edge AI component
The service employs multiple filtering techniques to optimize data transmission to AWS Cloud. Time-based filtering reduces data frequency through temporal sampling—for example, selecting one sample per second from data generated every 10ms. Statistical filtering applies basic calculations to data prior to transmission, such as computing mean values from multiple sources or averaging readings over time. With logical filtering, the service transmits data only when specific conditions are met, such as when values exceed defined thresholds or when representing minimum or maximum readings. You can implement these filtering methods individually or combine them to achieve optimal data reduction based on your specific requirements.
The Edge AI anomaly detection service enhances vehicle predictive maintenance and safety capabilities through advanced in-vehicle processing. The service receives periodic AI model updates pushed from AWS Cloud through OTA updates, ensuring that data filtration and selection processes leverage the latest AI learning. Using new generation automotive Systems-on-Chip (SoCs), the service’s AI engine performs transformer-based anomaly detection to monitor real-time data streams. This sophisticated monitoring identifies not only individual anomalies but also detects when data patterns across multiple sources deviate from expected behaviors, even if individual values remain within normal ranges. When the service detects pattern deviations, it automatically triggers bulk uploads, sending detailed data snapshots to AWS Cloud for comprehensive analysis.
On AWS
Figure 3 shows the entire eSync OTA solution running on AWS.
Figure 3: Solution architecture for eSync with the Edge AI component running on AWS
In each step, there is an aspect for training the AI model.
- Data Ingestion: Edge AI use AWS IoT Core to receive data from the vehicle fleet. This step ensures that data is accessible for subsequent processing and analysis.
- Data Storage: The data is stored in Amazon Timestream database, and in Amazon Simple Storage Service (Amazon S3) bucket for Extract, Transform, Load (ETL) and analytics.
- ETL Process: Following data ingestion and storage, Amazon Kinesis Data Streams and Amazon Data Firehose transform and prepare the data for analysis. To improve model training and analytics, the data undergoes structuring and cleansing. Presto on Amazon EMR makes it possible for domain experts to query and perform analytics on big data.
- Model Training: Amazon SageMaker plays a pivotal role in training a model of expected vehicle data and their relationships, using the recent data collected from the entire fleet. Individual vehicles then use this model to find anomalous data relations. Excelfore uses SageMaker Studio and the built-in transformers to fine-tune models, and SageMaker’s hyperparameter tuning and spot instances for productivity and cost benefits.
- Data Visualization: To facilitate insights into fleet performance and anomaly trends collected from the edge, Amazon QuickSight visualization tool is utilized. QuickSight reports help stakeholders to derive actionable insights from the processed data, promoting data-driven decision-making.
Key Benefits of Edge AI based anomaly detection
Selective data filtering, combined with expanded details on issues identified by AI anomaly detection, maximizes the value of the collected data by delivering several important advantages. The identification of anomalies and OTA updates serve as crucial tools for reducing the risk of vehicle failures and improving the reliability of the deployed fleet. The solution enables a continuous “learning loop” that allows for the ongoing refinement of vehicle functionality through the utilization of real-world data. Additionally, with its cloud-based design, the infrastructure effectively handles large fleets with millions of vehicles, offering the necessary scalability to accommodate increasing data volumes.
Conclusion
Continuous fleet improvement requires access to high-value vehicle data, such as vehicle component health and performance. However, uploading 1–2 terabytes daily is a challenge with typical 60-100 GB monthly upload limits. This blog explained how the Excelfore solution addresses the challenge using eDatX+AI data filtration, reducing data volume by 99.9% while selectively sending additional data when anomalies are detected. Using fleet-wide data, the AI model on AWS is constantly fine-tuned, automating and improving the process.
Excelfore eSync OTA deploys the fine-tuned AI models to the vehicles, completing the deploy-collect-analyze-improve-deploy loop for optimal fleet management. A growing list of customers, such as Plus AI, already benefit from Excelfore eSync OTA and eDatX solutions today. Excelfore has added this Edge AI feature as part of its commitment to help OEMs develop advanced software-defined vehicles. To learn more about eDatX+AI for eSync OTA, please visit the website here.
.
.
Excelfore – AWS Partner Spotlight
Excelfore is a differentiated AWS Software Partner and holder of the AWS Automotive Competency. Excelfore specializes in secure, AI-powered, cloud-native and standards-based connectivity and device management solutions for Software Defined Vehicles to collect data, provide Over-The-Air updates, and manage fleets. Excelfore is a technology leader and member of recognized industry organizations such as eSync Alliance, SOAFEE-SIG, ASAM, AVNU and AUTOSAR.