From Raw Sensor to Business Signal: An Azure Databricks Framework for Diesel Engine Performance Analytics Across Global Fleet Operations

Abstract

A framework for the scalable analytics of diesel engine performance across global fleet operations is presented. Key performance metrics from raw sensor data are defined, demonstrating a clear link to business objectives such as fuel consumption and maintenance tackling times. Modelling such metrics requires appropriate sensor data (estimated here at 25 to 35 signals) underpinned by data processing and feature engineering activities, supported by both real-time and batch solutions. Performance from the analytics framework is achieved through a breadth of modelling approaches: time-series modelling  of performance drift, classification modelling of engine failure avoidance, and popular machine-learning performance-influence exploration. Supporting infrastructure is within the Azure Databricks ecosystem, alongside a dialogue with fleet operations through an Microsoft Azure application. The solution, wholly designed and implemented by a single data scientist, provides all components required for production-ready analytics. Results are being deployed within a multi-Cloud platform that extends from raw sensor data through to business summary dashboards, encompassing a broad range of analytics beyond those for diesel engines.