Case Study
Reducing rework in axle manufacturing
Objectives
- Locate source(s) of failure in axle assemblies
- Reduce failure and rework rates using automotive manufacturing data
Challenge
- Assemblies require >200 measurements across 20 different operations to produce one unit
- Multiple sources contributed to variance in component selection
Key Results
Reduced failure and rework rate by
65%
Significant improvements to manufacturing throughput and cost optimization
Background
A major Tier 1 supplier was looking to leverage machine learning techniques on production assembly data to reduce the rate of product fallout and associated rework for axle assemblies. A company-wide initiative to deploy such advanced manufacturing solutions resulted in an engagement with Acerta, the only SaaS provider that demonstrated an extensive history of automotive manufacturing use cases for machine learning.
Several production facilities were considered for the initial deployment, with LinePulse ultimately being utilized globally on multiple lines in several manufacturing facilities to provide insights across the client’s supply chain.
The Problem
Just one of the client’s production lines consists of more than 20 different operations which collectively generate over 200 measurements per assembled unit. Given the number of available measurements that can influence each unit, narrowing down the sources of failures was critical. This difficulty was compounded by the existence of multiple failure modes, each of which could have a different underlying set of causes.
Solution Process
Acerta initially assisted with qualitatively and quantitatively describing the failure mode by looking to capture information on any complex interactions or information that might be present outside the standard arithmetic methods. LinePulse identified several key measurements from the manufacturing process that most reliably predicted backlash failures. This enabled the client to narrow down the list of likely causes of the failures—including previously unsuspected relationships resulting from measurements across operations— without the usual manual effort involved in sifting through large volumes of production data.
These early results helped build confidence in a more progressive AI/ML approach, that could integrate with the client’s network layer and down to the PLC, whereby a model built on the insights from the Linepulse platform could be deployed. As someone familiar with axle or PTU assemblies might know, setting the correct backlash is critical to function of the assembly but is typically done using shims calculated by considering the relevant component tolerance stack. In a perfect world this calculation should always be able to set the nominal backlash and provide the correct preload, but the reality is that this is not the case. Acerta was able to prove out and deploy a decision point solution that replaced the tolerance stack calculation with ML models that would pull an assembly’s entire build record, as well as previous assemblies, leveraging all information to provide a more accurate output.
Working through progressively production ready trials through the Linepulse Platform, adapted and customized to fit the plant network infrastructure, Acerta was able implement the solution at a production scale across two assembly lines in the facility.
Results
The impacts on cost optimization and throughput observed in initial deployments of Acerta’s Anomaly Detection and Capability Metrics modules have led the customer to utilize LinePulse across facilities globally and prioritize investment into these Industry 4.0 and AI initiatives. The implemented system achieved a 65% reduction in failure and rework rates, leading to a significant improvement in first time yield and cost reductions for the client.