Tag: Learning

By Keith E. Greenberg, SAP

Automotive manufacturer Mahindra Heavy Engines Limited (MHEL) has been building powerful diesel engines for more than 70 years. But changing market demands forced this venerable company to face a 21^st century dilemma.

MHEL needed an extended quality testing program for internal combustion engines in order to reduce cost while accelerating the product manufacturing lifecycle.  And there was no time to waste. 

But because of the technical resources required, and the fact that the data was staged in multiple stand-alone servers, the process was slow and costs high.

So, the Mumbai-based company took a radical step: replace its outmoded physical methods with virtual testing – using artificial intelligence/machine learning (AI/ML) while incorporating data from a variety of sources.

How it’s done

To understand MHEL’s challenges and objectives, we need to take a brief glimpse at how things work there.

Currently, quality testing accounts for one percent of engine manufacturing costs. In the final phase of quality testing, the engines undergo what are known as “cold” and “hot/load” tests to identify defects and ensure quality.

What that means: in a cold test, the engine’s crankshaft is rotated with an electric motor, while software analyzes data from different sensors. 

• Tests take approximately 140 seconds for each engine.
• Engines that fail then undergo hot/load testing – which requires the engine to be fired and take two to three minutes.
• Then come the load tests, which can go as long as 12 minutes.

Many engines don’t need the hot/load tests. MHEL’s challenge was eliminating these unnecessary steps without compromising quality.

Driving change

To hear company representatives tell it, MHEL has thrived because of unconventional thinking and the innovative ways it has utilized resources. 

Therefore, the logic went, why couldn’t MHEL rely on ML to validate engine quality, eliminating the need for avoidable tests for those engines that already met hundreds of predefined parameters?

To make that happen, a predictive quality ML model had to be designed to analyze test results and other data to determine engine performance and quality. This tool would identify the likelihood of oil leaks and other flaws generally detected during hot/load testing. 

Improvising possibilities

MHEL turned to SAP to manage business operations. Drawing on solutions created through the SAP Integration Suite, the new model eased complexity by unifying data from an array of sources – calculating in such factors as engine suppliers, in-house manufactured parts, engine assembly, and user-plant defects.

This wealth of information, combined with specific parameters and cold test results, allowed each engine to be positively classified as “Further Hot Test Required,” “Further Load Test Required,” or “No Further Test Required.”  Remember: prior to the introduction of the new model, even engines that didn’t need additional testing underwent the extra steps — costing MHEL both time and money.

“This first-of-its-kind use