Field Intelligence guest column
Rolls-Royce Power Systems’ AI journey started in 2019 at a South Carolina facility producing large diesel engines for use in generator sets, naval and marine applications and military vehicles.
Upon completion of the assembly process, each engine is subjected to rigorous testing. During this testing process, subtle indications of a pending problem may go unnoticed by even the most experienced operators—potentially leading to a catastrophic failure during the test or after the engine is put into service.
These failures caused scrap, rework, delayed shipments, backlogs and warranty claims, costing Rolls-Royce millions of dollars annually.
The problem was not a lack of data; the plant had been collecting process data for years. But the plant only used the data for follow-up root cause analysis after failures.
Over 50 billion data points were collected from multiple systems for analysis.
Using Delta Bravo’s technology, Rolls-Royce was able to, within 48 hours, reduce the dataset to 2 billion critical data points. A gap was identified in data collection frequency. Increasing collection frequency from .5 Hz to 10 Hz at a critical process juncture enabled development of an accurate machine learning (ML) model.
Detailed correlation analysis identified the metrics that most influenced failure and baselines were established for each.
With each trend or anomaly observed, the ML model was tuned and retrained. Within a few weeks, the model identified a group of at-risk engines by serial number. These engines had an above-average probability of
experiencing a problem during the factory quality control test procedure or in the field.
By correlating the test data to actual product failures, the report accurately identified over 80 percent of the engine problems over a period of several years.
The next step was fitting the new predictive capability into existing operator processes happening in real time. We integrated with Rolls-Royce’s existing test cell software via API, enabling the model to automatically stop the test and give the operator real time feedback in the system to which they were already accustomed. Operators received detailed feedback on what to adjust prior to resuming the test, and zero operator training was required.
Within 45 days, the model was able to predict failures 30 minutes in advance, with zero false positives. Within 90 days, the AI solution saved multiple engines, producing a 10x project ROI.
The model was extended from the test cells to the field, predicting low-hour field failures, reducing associated warranty claims and maintaining positive brand loyalty with customers.
In recognition of this project’s use of AI to predict and address quality issues, the manufacturer was named innovator of the year by the South Carolina Manufacturing Extension Partnership.
Rolls-Royce envisions a future built around better decisions at each process, operational and management level.
Significant investments have been made in equipment and systems that collect data for each step of the process.
Cross-functional resources from operations, finance and IT work together to identify the problem, quantify the impact of solving it and access the data in a secure and compliant way. This shared vision and priority was what it took to get the right people in the room not just to identify the problem, but to validate, integrate and scale the solution quickly.