Reducing casting defects digitally: Al slashes iron foundry’s scrap by 50%

One of the largest in the southern hemisphere, this South African-based foundry has an annual melting capacity of 110,000 tonnes and produces over 75,000 tonnes of high-quality cast iron automotive components every year.

Back in 2017, scrap, rework and shipping castings with subsurface defects were costing the foundry dearly. Although it invested in Industry 4.0 systems, the operations team found that optimising its entire casting process was too complex for manual analytics.

Instead, it turned to artificial intelligence (AI) and deployed the Monitizer | PRESCRIBE application. Monitizer | PRESCRIBE is the product of Norican Group’s partnership with DataProphet. Combining Norican’s deep foundry experience with DataProphet’s award-winning expertise in manufacturing artificial intelligence and data science, this packaged service quickly delivers insight that significantly improves foundry performance over both the short and – as the system keeps learning and improving the process – the long term.

Deploying Monitizer | PRESCRIBE gave the foundry full control over process optimisation, with real-time machine settings that helped cut defects by 50% in the first month of deployment, pushing defect rates in shipped castings close to zero after just three months and saving the foundry over $100k every month.

Monitizer® | PRESCRIBE helped a green-sand iron foundry to massively reduce scrap and save over $100,000 every month

The Challenge

Stopping internal and external defects

This foundry’s vision is to be the best in the world. But to compete globally, it needed to optimise production to reduce the cost of scrap and achieve stable predictable output.

In particular, it wanted to avoid shipping components like engine blocks internationally that then had to be discarded due to internal defects that only became apparent after machining. Though the foundry’s scrap rate was similar to the industry average, long-distance shipping to customers added substantially to the cost of scrap.

Lowering the number of defective engine blocks shipped by reducing the internal and external defect and scrap rates became the main objective. In turn, this would increase overall production yield and volume as well as reducing the delays caused by reworking.

The foundry understood that Industry 4.0 systems and techniques were the route to achieving its quality goals. It had already upgraded digitisation and data acquisition across all stages of production, using production and quality data to help plant engineers and managers to make better, more informed decisions.

However, trying to optimise an entire foundry line can involve hundreds or even thousands of variables across all the different process stages, with large numbers of non-linear causal relationships between them. Making sense of this extreme complexity appeared impossible, with process engineers faced with an avalanche of information describing millions of possible combinations of plant states. As data-driven improvements plateaued and innovation stagnated, foundry management decided to explore artificial intelligence.


Powered by AI

To get the answers it needed, the foundry chose to use Monitizer | PRESCRIBE. This AI tool employs automation and the power of cloud computing to tackle large sets of data and extremely complex analyses head on. Where traditional manual statistical methods struggle, AI automatically examines historical data to learn how parameters like sand grain size and moisture content, melt pouring speed or inoculation rate influence each other – and affect final casting quality.

Monitizer | PRESCRIBE calculates which combination of machine and material settings will produce minimum defects, then provides real-time prescriptions (recommendations) on the settings required to maintain stable, maximum quality production. Customers receive reports and advice through a user-friendly web-based front end. Because Monitizer | PRESCRIBE is a cloud-hosted, Software-as-a-Service application that integrates with existing infrastructure, there is no need for any new IT investment.

Creating an accurate and reliable data set that described how the line performed at different times was just as important as the subsequent analysis, so the project began with extracting, transforming and loading (“ETL”) historical production data into a data warehouse. That data was sourced right across the entire plant: multiple departments, the foundry’s central SCADA systems and other industrial control systems.

The data arrived in various formats, from handwritten records and Excel files to Access database data and CSV sources. This was digitally transformed to create a historical view of 15 months of production that contained all the important process features for the castings produced.

Once digitised, the process and casting quality data was ingested into Monitizer | PRESCRIBE solution.. Through advanced supervised and unsupervised machine learning, the AI automatically discovered the optimal operating regime for the foundry’s entire production process. The machine learning neural network model calculated and established the correlation and interaction between approximately 1000 plant-wide parameters.

Operators, engineers and foundry managers across the entire operation now have their own customised view of the latest recommended parameters via an interactive web-based interface. Prescriptions are prioritised in each report to ensure that the most important changes are made first in every part of the line. Though control of the line remains manual, the AI’s real-time supervision means the plant operates within, or very near, the identified optimum regions, with minor variability remaining due to uncontrollable variables.



The fast track to digital excellence

Impressive results arrived quickly. In the first two days, applying the AI prescriptions drove internal scrap down from 5-6% to 1% and casting reworking fell from 10-15% to 8%. In the first month of AI deployment, the foundry halved its scrap rate and, within the first three months, achieved an external scrap rate below 0.1%.

For the first time in the history of the company, defects came down to zero for periods of up to three months. Since deploying Monitizer | PRESCRIBE, the foundry saves around $100k per month and has simultaneously increased yield: the foundry achieved record production outputs in 2018 and 2019.

The low external defect rate has also cut wasted production energy and the unnecessary transportation of large defective engine blocks, saving an estimated 135kg of carbon dioxide emissions for each defective block not shipped.

DataProphet is a leader in AI that enables manufacturers to step towards autonomous manufacturing. Their AI-as-a-service proactively prescribes changes to plant control plans to continuously optimize production without the expert human analysis that is typically required.

As recognized by the World Economic Forum, DataProphet applications have helped customers around the world experience a significant and practical impact on the factory floor, reducing the cost of nonquality by an average of 40 percent.

In the past we have been able to achieve similar results, but we had absolutely no clue what we did to achieve the good result. This time round we have a fairly good idea of what we did to achieve the result. Thanks for your tremendous efforts and help to take us to the next level.

Foundry's CEO