Viaduct’s unified data model links vehicle telematics to parts, service, and repair events in a structured and consistent way – making it easy to generate, codify, discover, and enrich knowledge about any failure event throughout the vehicle lifecycle.
- Schematization and labeling.
Automatic mapping of raw data to target schemas optimized for machine learning use cases (e.g., organizing unstructured dealer notes for downstream anomaly detection)
- Data validation.
A pipeline that automatically detects data quality issues (e.g., odometer readings that don’t increase monotonically, geospatial errors, misconfigured DTCs that have never fired) and reconciles conflicting readings across systems (e.g., disparities between odometer readings in warranty claims vs. telematics data) - Features and factors.
Inference of structural relationships between heterogeneous, high-dimensional data by Viaduct’s TSI Engine – generating a reusable library of features (including short-term and long-term varying factors – e.g., recent spikes in oil temperature, along with a gradual decline in fuel efficiency over many months). Codification of factors (human-explainable features) is a critical input to root causing and model interpretability/trust.
Viaduct maintains a latent representation of vehicle health as a set of relationships between factors derived from multiple data sources:
- Patterns in raw sensor data
- Natural language processing to identify patterns across technician notes in inspection reports (e.g., “ticking noise in engine,” “tick at regular intervals”)
- Time-varying correlations – e.g., a population of vehicles that experienced DTC [X], followed by sensor pattern [Y]
Deviations from known relationship patterns signal potential anomalies. The platform uses the Temporal Structural Inference Engine’s
proprietary combinatorial optimization algorithm to efficiently identify vehicle subpopulations (across build data and operating characteristics) with high incidence of a given quality anomaly. (E.g., we identify an anomaly from DTC clusters, sensor patterns, and NLP on technical notes from warranty claims – and then isolate the affected subpopulation to vehicles with a certain engine component assembled in one plant between May and August 2022, driven primarily above 4,000 feet in elevation.) Viaduct projects the future impact of these issues through survival mode analysis.
Viaduct predicts failures on the individual vehicle level over any time-, odometer-, or utilization-based horizon. Viaduct’s Temporal Structure Inference (TSI) Engine identifies related components and subsystems – enabling parameter-sharing to improve model performance without risk of overfitting.
- Robustness.
Predictions harness a library of features built specifically for connected vehicle failure use cases – including features derived from the interactions of multiple sensors, failure indicators, and/or warranty claims (e.g.,harsh braking). Viaduct’s proprietary joint regularization approach, embedded within the TSI Engine, enables model performance to improve by sharing parameters – without risk of overfitting, even in the case of extremely infrequent failure events. - Efficiency.
Viaduct maintains a latent-state representation of every vehicle’s health that gets continuously updated as new information becomes available, making the process of updating and maintaining models fast and easy. - Transparency.
All ML models are built for interpretability – no black-box features without a human-explainable definition.
Viaduct makes it easy for organizations to turn insights from connected vehicle data into customized service plans.
Business operators use Viaduct's rules framework to define alert thresholds, service schedules, and maintenance actions based on analytic insights. And to support getting insights into the field, Viaduct offers:
- Out-of-the-box integrations for both data ingestion and connectivity todownstream systems like CMMS’s and maintenance plans.
- A REST API-based architecture enabling Viaduct predictions to enrich any operational application.
- Human-explainable modeling outputs that make it easy for teams to socialize emerging quality issues, root causes, and customized maintenance recommendations across the organization.
Viaduct's hands-on service model helps teams harness the expertise and best practices of automotive industry and machine learning veterans.
