Logo of Acerta
Book a demo

Case Study

Cutting Transmission Warranty Costs by 30%

Objectives

Reduce warranty claims and accelerate root cause analysis for faulty transmissions by enhancing end-of-line (EOL) testing with machine learning

Challenge

The client had very limited data—only 100 transmissions and no recorded failures—and needed real-time test results to catch hidden defects before shipping.

Key Results

  • Reduced signals requiring root cause analysis by 99.8%.
  • Reduced warranty claim costs by up to 30%.
Vehicle transmision

99.8%

Fewer signals to investigate

30%

Reduction in warranty costs

Background

A leading Tier-1 transmission supplier aimed to reduce warranty claims by detecting manufacturing defects using production and end-of-line test data. Each EOL test involved over 100 steps covering various performance-based quality assessments. The supplier’s existing Statistical Process Control (SPC) system analyzed only 10% of collected data, leaving engineers to manually inspect all signals in cases of potential failure—an inefficient and time-consuming process.

Problem

Sparse and Unlabeled Data

The client supplied training data from 100 units, all of which passed EOL testing and had no reported warranty claims. This presented a challenge: how to train an anomaly detection model with no labeled failures.

Manual Inspection Burden

The existing system flagged only clear, extreme failures, requiring engineers to manually inspect thousands of signals to detect more subtle issues.

Real-Time Requirement

The client needed the new solution to integrate into their EOL process and provide real-time results without delaying production.

Solution

Targeted LinePulse Deployment

Acerta’s team began by collaborating closely with the client’s manufacturing and data collection teams. This allowed for intelligent feature engineering tailored to the client’s processes.

Through non-polynomial feature extraction—identified as valuable based on past use cases—Acerta reduced feature dimensionality to ensure the model focused on the most impactful signals.

Unsupervised Machine Learning Approach

Given the lack of labeled failure data, Acerta deployed LinePulse’s unsupervised learning algorithms, which calculated an abnormality score for each transmission based on:

  • Single signal behaviors
  • Multi-signal relationships
  • Signal behavior across multiple test steps

The score leveraged ensemble models’ reconstruction errors to highlight deviations from normal behavior.

Real-Time Monitoring & Root Cause Acceleration

LinePulse ranked transmissions based on abnormality scores and pinpointed the least explainable signals contributing to anomalies. For example, it identified a causal link between pressure delays and rotation delays—a relationship previously undetected by the SPC system.

The number of signals requiring manual review dropped from 4,000 per transmission to just 10, significantly accelerating root cause analysis.

Results

  • 99.8% Reduction in Signals Requiring RCA:
    Engineers could focus only on the most abnormal signals, streamlining investigations.

  • 30% Reduction in Warranty Claim Costs:
    Improved detection at the EOL reduced the number of defective transmissions reaching customers.

  • Multi-Signal Failure Detection:
    Unlike the existing SPC program, LinePulse detected subtle, multi-signal issues and provided a high-confidence abnormality score without requiring extreme value thresholds.

The client successfully integrated LinePulse into their EOL testing process, benefiting from real-time detection and long-term cost savings.

Case Studies

Proven Impact