# Architecture overview
<a name="pm-architecture-overview"></a>

The Predictive Maintenance solution follows a multi-stage architecture that processes vehicle telemetry data through parallel prediction pipelines.

![Predictive Maintenance Architecture](http://docs.aws.amazon.com/guidance/latest/automotive-data-platform-on-aws/images/predictive.png)


## High-Level Architecture
<a name="high-level-architecture"></a>

The architecture consists of:

1.  **Data Source Layer**: Amazon Redshift cluster containing vehicle telemetry data

1.  **ETL Processing Layer**: AWS Glue jobs that transform and prepare data hourly

1.  **Prediction Layer**: Dual-path processing with ML and filter-based approaches

1.  **Alert Layer**: Consolidation and delivery of maintenance alerts

1.  **Storage Layer**: Amazon S3 for intermediate data and Amazon DynamoDB for alert state

## Data Flow
<a name="data-flow"></a>

The solution processes data through the following stages:

### Stage 1: Data Ingestion
<a name="stage-1-data-ingestion"></a>
+ Telemetry data is accessed from Amazon Redshift via Datashare or S3 unload
+ Root ETL pipeline runs hourly to extract new data
+ Data is transformed into standardized formats and stored in S3

### Stage 2: Parallel Processing
<a name="stage-2-parallel-processing"></a>

Two independent pipelines process the data:

 **Machine Learning Pipeline:** \* ML ETL prepares features from historical data \* Training pipeline updates models weekly using Amazon SageMaker \* Inference pipeline runs batch predictions on new data \* Anomaly scores are generated for each vehicle/tire combination

 **Filter-Based Pipeline:** \* Statistical filters analyze pressure trends \* Leak rates are calculated using time-series analysis \* Alerts are generated when thresholds are exceeded

### Stage 3: Alert Consolidation
<a name="stage-3-alert-consolidation"></a>
+ Results from both pipelines are merged
+ Duplicate alerts are deduplicated
+ Severity levels are assigned based on leak rates
+ Alert status is tracked in DynamoDB

### Stage 4: Alert Delivery
<a name="stage-4-alert-delivery"></a>
+ Alerts are formatted for downstream systems
+ Integration APIs provide real-time access to alert data
+ Historical alert data is maintained for analysis

## Key AWS Services
<a name="key-aws-services"></a>

The solution leverages the following AWS services:
+  **Amazon Redshift**: Source data warehouse for vehicle telemetry
+  **AWS Glue**: Serverless ETL for data transformation
+  **Amazon SageMaker**: ML model training and batch inference
+  **Amazon S3**: Data lake storage for intermediate and processed data
+  **Amazon DynamoDB**: Alert state management and tracking
+  **AWS Lambda**: Event-driven processing and orchestration
+  **Amazon EventBridge**: Scheduling and workflow coordination
+  **AWS Step Functions**: ML pipeline orchestration