The BMW Group manages 1.3 PB of vision and sound analytics data to drive AI innovation using AWS

BMW’s portfolio included a number of distinct products tailored for audio, image data management, and computer vision. The challenge lay in that audio and image data were underpinned by disparate technologies and methodologies. BMW wanted to unify these systems into a single application framework, modernizing and standardizing its architecture so that it could use this data for acoustic- and image-related products. Building on its existing AWS relationship, BMW chose to use AWS to integrate this data to uncover new insights.

To achieve its standardization goal, BMW developed the VSAS on AWS. VSAS started as a centralized image storage solution for production quality inspections but quickly evolved to include audio data storage and computer vision–driven data management for efficient data labeling and AI compatibility. This comprehensive database, containing over 19 billion metadata elements, helps employees to locate manufacturing parts and perform advanced analytics efficiently.

BMW understands the critical role that images play in today’s visual-centric world, and its goal is to empower business departments to securely store, organize, and retrieve their image assets with ease in near real time. The potential for using this data extends beyond immediate operational enhancements. It provides a foundation for developing models that can anticipate quality and production issues before they occur. The integration of AI into VSAS not only supports current manufacturing demands but also paves the way for future innovations, making sure that BMW remains at the forefront of the automotive industry. For training and building its AI models, BMW uses AWS Graviton processors, which offer the best price performance for a broad range of workloads.

BMW uses Amazon OpenSearch Service, an AWS-managed service that lets users run and scale OpenSearch clusters, to effortlessly handle massive volumes of image metadata across multiple resources and source clients, with infrastructure that can process inspection data, archive metadata, and image records simultaneously without performance degradation. The query architecture provides exceptional flexibility through search metadata functions, facilitating complex filtering, sorting, and aggregation queries across different object types—images, metadata, inspections—while maintaining consistent response times even with large datasets.

To make sure that VSAS would be compatible with its existing camera systems, BMW used AWS Transfer Family to easily manage and share data, simplifying file migration without disrupting existing connections. This approach reduced complexity and costs while modernizing the infrastructure. The company also uses Amazon Elastic Container Service (Amazon ECS), a fully managed container orchestration service, for management and scaling of containerized applications and services.

During development, BMW worked closely alongside AWS to optimize VSAS, making significant improvements. For example, the team identified that initial authentication time was up to 5 seconds during early testing. Using AWS, they reduced the time to authenticate data to under 1 second, reducing total authentication time by 80 percent. This enhancement improved VSAS’s data ingestion capabilities, enhancing its scalability, reliability, and responsiveness.

With access to AWS infrastructure, BMW avoids lengthy turnaround cycles for procuring infrastructure, helping it to be more agile and scale rapidly. The stability of AWS infrastructure means that BMW doesn't need to worry about operational issues.

Initially, VSAS deployed to three production sites, processing approximately 300,000 images daily and supporting 43 TB of data. The solution grew rapidly, doubling in size within the first year and expanding to 16 production sites over 3 years. It now processes an average of 1.3 million images and audio files daily, managing 1.3 PB of data.

By using Amazon S3, the solution efficiently scales to manage growing data volumes. AWS has made it possible for BMW to expand seamlessly even beyond the company’s initial growth expectations. The VSAS team is relieved of concerns about running out of storage or restricting business operations; using AWS services provides the scalability needed to accommodate large data volumes and support ongoing growth.

While increasing data storage, the company achieved a cost reduction of over 60 percent using Amazon S3 Intelligent-Tiering storage classes, which automatically optimize storage costs. BMW further reduced costs by 63 percent for its computer vision solution by using AWS Graviton processors, which provide the best price performance for cloud workloads running on Amazon EC2. Transitioning to Graviton-based GPUs for running machine learning tasks, especially training neural networks related to computer vision, made it possible for BMW to reduce its compute costs while delivering the same compute output as the previous GPUs.

The cost-efficiencies of the VSAS solution help BMW to capture more images during production, facilitating AI model development for targeted quality checks. The company uses the data stored in VSAS to build AI models for predictive maintenance and quality management, using Supervisely software on Amazon EC2 to enhance the reliability of its production infrastructure and streamline maintenance workflows. The computer vision solution predicts potential rail failures, which reduces costs, minimizes downtime, and improves overall equipment effectiveness across BMW’s manufacturing facilities.

For quality management, BMW developed the Acoustic Analytics application to automatically capture and analyze vehicle sounds to improve engine performance and in-cabin comfort throughout the vehicle’s life cycle. This automated approach is valuable for testing development vehicles, where no driver is present to detect anomalous sounds. VSAS stores and labels the audio data, then uses the data to create and train new models, improving diagnostic accuracy while reducing the need for human intervention.

VSAS offers a powerful self-service solution through its intuitive acoustic analytics model builder, empowering users to create custom noise detection AI models for diverse use cases. This approach facilitates seamless data acquisition, with users able to gather any relevant audio data and precisely label noises. These audio files can then be efficiently organized into an audio collection, which serves as the foundation for training new acoustic analytics models. With the click of a button, users can use these collections to train their own specialized AI models, all of which are built and trained using the robust infrastructure of AWS. These newly trained models can then be deployed to perform predictions on other audio files. These predictions are powered by Amazon SageMaker—which delivers an integrated experience for analytics and AI—for efficient and scalable inference.

By centralizing its audio and image storage on AWS, BMW created a flexible system that provides consistent quality management across all its global production sites while remaining agile and cost-effective.

BMW strives to develop a safer, more efficient, and more pleasurable driving experience for its customers. One key area of focus in achieving these goals is data-driven vehicle development. By analyzing vast amounts of data, the company gains insights into the performance and behavior of its vehicles, leading to more informed design decisions. For example, by capturing and analyzing acoustic data from a vehicle, BMW gains insights into everything from engine performance to in-cabin comfort.

The company continues expanding VSAS into new business units so that they can build their own computer vision and acoustic analytics models. This self-service approach equips users to swiftly train, evaluate, and deploy AI models, significantly reducing the time and effort needed to develop solutions. The system provides a comprehensive data analysis solution, supports AI development, and optimizes costs and operational efficiency, making it possible for the team dedicate 80–90 percent of resources to innovation. Using training instances from Amazon SageMaker, BMW decreased the model training time for various use cases from days to hours. The flexibility of these instances also helps the company automate the training process, accelerating the delivery of value to the business.

BMW’s use of AWS services helps the company to quickly scale its infrastructure based on project requirements, facilitating efficient management of large data volumes without traditional limitations. Without using AWS, BMW would need a substantial infrastructure team to handle the data volume. Using a suite of AWS services, BMW can now concentrate on delivering business value and creating innovative products for its customers.