Deep Learning Optimization for Engine Block Manufacturing | a Case Study

Our client is a world-leading foundry with a presence in over forty countries across two continents. As the world’s largest manufacturer of cast iron engine blocks and cylinder heads—they supply globally for the commercial, agricultural, and heavy vehicles automotive industry.

Our customer works with cast iron alloys like gray and ductile iron, but also with next-generation compacted graphite iron (CGI), which it has mastered the utilization of. Their one industrial plant alone has a melt capacity of over 400 thousand metric tonnes annually.

Seeing the potential value of AI to further reduce baseline scrap, this leading manufacturer acquired DataProphet’s services through a partnership with Kaitronn, our representative for Latin America.

THE CHALLENGE—REDUCE BASELINE SCRAP

Despite the highly sophisticated equipment of advanced foundries, an extra competitive advantage with respect to production capacity can be elusive. This, in large part, is due to the sheer weight of information process engineers now have access to, which they must parse in order to choose an ideal from millions of possible plant setpoint combinations.

The objective of our installation was to discover an optimal production paradigm and to generate enhanced control plans for plant operators. If successful, the executed control plans would deliver more optimized production output for the products on the line. The result would be consistently less generation of scrap.

Tupy-case-study-diagram

TECHNICAL CONSIDERATIONS FOR THE MANUFACTURING CONTEXT

External defect reporting is, by nature, slow and unreliable. As such, we would need to account for the source of as many internal defects as possible—this way, optimal prescriptions could proactively reduce them. Additionally, we would have to factor in the subset of defects that could be either process-related or operation-related. The goal was to optimize the line as quickly as possible while not interfering with production.

THE SOLUTION | TRACING THE PROCESS

DataProphet PRESCRIBE AI algorithms learn the interactions and interdependencies between industrial processes and their parameters. The AI works by sculpting low-dimensional representations of complex, high-dimensional production data. The final training set consisted of over 150 thousand engine blocks, each with greater than 70 process parameters.

THE SOLUTION | PRESCRIBING THE BEST-OF-BEST (BOB) REGION

DataProphet’s algorithms discovered a region that corresponded to historical production—where both a low defect rate (quality) and a sufficient number of blocks (density) were observed. Viable BOB regions having been identified according to the agreed parameters for the two products on the line—a non-disruptive path to an optimal control plan was laid.

DOWNLOAD THE CASE STUDY

WHY CHOOSE DATAPROPHET AS YOUR AI PARTNER

Too often, AI-as-a-Service providers equate research-based methodologies with one-size-fits-all industry solutions. DataProphet works closely with partners to gain an understanding of their unique set of production problems. We adopt AI-driven manufacturing solutions to the production dynamics of your factory floor, incorporating our proven, industrial process expertise, but taking our cues from contextualized feedback. We understand that most impact is achieved by building machine-learning models to mesh with your production lines' realities and the assets that drive them. Our partnerships are guaranteed win-win.