Panasonic Says Its Aircraft AI Cut Diagnostic Investigations From Hours to Minutes

Panasonic Avionics has deployed an AI system that turns fleet monitoring into parallel fault investigations and incident routing. The system can create and prioritize critical alerts, while engineers approve operationally significant remediation. Panasonic says targeted use cases improved operational efficiency by 20–40%, and its internal testing reduced investigations from hours to minutes.

One fleet, many diagnostic baselines

Panasonic Avionics built the system with AWS and the AWS Generative AI Innovation Center for in-flight entertainment and connectivity operations. Individual deployments have tailored configurations and distinct log patterns, making fleet-wide assessment difficult. Engineers previously correlated logs, metrics, and ticketing data by hand, a process that could take hours and required deep system knowledge.

Detection feeds three parallel checks

The workflow standardizes raw fleet data into service metrics in an Amazon S3 lakehouse. A domain ontology, a shared vocabulary for fleet entities and their relationships, links performance metrics to configuration metadata and ticketing information so diverse deployments can be compared. A Trend Analyzer then evaluates performance and degradation signals for anomalies, including patterns shared by a configuration variant.

• The Correlation Analyzer looks for recurring patterns across shared configurations to assess whether an issue is isolated or systemic.
• System Checks compare metadata and service status with ticketing workflows, including whether known maintenance explains the behavior.
• The Log Analyzer matches current patterns against previously identified failure modes using rules and pattern-based detection.

Amazon SageMaker uses LangGraph to orchestrate those investigations in parallel. The system also retrieves similar historical incidents and resolution artifacts from Amazon RDS with pgvector, including cases where symptoms are not exact matches. Anthropic Claude on Amazon Bedrock converts those findings into reports with root-cause hypotheses, fleet-impact analysis, and prioritized recommendations.

Automation routes critical work; engineers retain the decision

The system produces daily diagnostic reports for Panasonic’s active fleet. For critical findings, it creates alerts, prioritizes incidents, and routes them to the relevant engineering teams. Recommendations use retrieved operating data and historical incidents, undergo deterministic business-rule validation, and require human approval for operationally significant remediation.

Panasonic says it limits large-language-model use to summarization and error reasoning, while retaining decision traces for auditability and improvement. The architecture is a specific implementation; AWS and Panasonic say outcomes can vary with data, operating context, and configuration.