Case Study

Diagnosing engine failures with 93% accuracy

Objectives

Help the client keep vehicles running smoothly by using AI to predict engine failures and show why they happen, across different engine types.

Challenge

There were different types of engines and tests, making it hard to spot failures—and results needed to be delivered in real time with 80% accuracy.

Key Results

Achieved 93.3% accuracy in failure predictions.
Identified suspect test signals in 100% of failure cases.
Enabled technicians to catch more failures and reduce investigation time.

93%

Reduction in failure misdiagnosis

100%

Suspect signals identified

Background

A leading Tier-1 engine supplier needed to enhance its vehicle diagnostics system to minimize breakdowns and improve repair times. Their current approach relied on aggregated diagnostic trouble codes (DTCs) via the on-board diagnostics port, which was insufficient for capturing different engine failure modes. Additionally, root cause analysis was manual, slow, and resource-intensive.

Problem

Diverse Platforms & Profiles: Four engine platforms with 4-5 test profiles each complicated the creation of a machine learning model to solve the problem.
Multiple Failure Modes: Existing diagnostics tools struggled to capture four specific engine failure types.
Real-Time Needs & High Accuracy: The customer required predictive models with 80%+ accuracy, so they could use the results in real-time to improve servicing speed.

Solution

Tailored Models for Each Engine Type

Acerta’s team began by closely examining the client’s engine test data. Recognizing that each of the four engine platforms had unique behaviors and test profiles, they chose to build separate machine learning models for each platform. This allowed the models to be customized to the specific characteristics of each engine type, rather than relying on a single model to cover all cases.

Combining Two AI Techniques

To achieve the client’s goals of both predicting engine failures and identifying their root causes, Acerta applied two different AI techniques. Classification models were used to predict whether an engine was likely to fail, while anomaly detection models highlighted unusual patterns in the engine’s test signals. This combination ensured that not only could failures be predicted, but the reasons behind those failures could be clearly understood.

Real-Time Results Delivered to Engineers

Acerta developed a streamlined system to process the engine test signals and deliver results to the client’s engineers in real time. The LinePulse Advanced Anomaly Detection dashboard provided clear, actionable information—including engine ID, test details, pass/fail status, abnormality scores, and specific areas in the signal data where issues occurred.

Results

Achieved 93.3% accuracy in predicting engine failures, surpassing the client’s 80% target.
Identified 100% of suspect signals and anomalous regions in engine test data.
Enabled technicians to detect failures earlier and reduce time spent on root cause investigations.
Improved vehicle servicing speed and reliability by providing real-time diagnostic insights.